Intelligent analysis and prediction method and system for multi-modal data of sudden cardiac death high-risk group

Abstract

The invention provides a sudden cardiac death high-risk group multi-modal data intelligent analysis and prediction method and system, and relates to the technical field of medical health monitoring, and the method comprises the steps: collecting multi-modal data, projecting an electrocardiogram to a three-dimensional coordinate system, extracting potential trajectory features, and calculating a space-time discrete index; performing dynamic response analysis on the blood pressure data, and matching with a space-time discrete index to calculate a cardiovascular coupling coefficient; and performing segmented correction according to the motion state data, monitoring a coupling coefficient deviation condition in real time, calculating a risk degree and generating an early warning. According to the method, heart abnormity can be found in advance, and sudden cardiac death can be effectively prevented.

Description

Technical Field

[0001] The present invention relates to medical health monitoring technology, and in particular to a method and system for intelligent analysis and prediction of multimodal data of people at high risk of sudden cardiac death. Background Art

[0002] Sudden cardiac death is a fatal cardiovascular event characterized by rapid onset and severe severity, making it a major public health threat to human health. Currently, clinical risk assessment relies primarily on single physiological indicators such as electrocardiograms and blood pressure, which fail to fully reflect the dynamic coupling state of the cardiovascular system, resulting in inaccurate early warnings.

[0003] Existing sudden cardiac death warning methods often overlook the impact of exercise load on cardiovascular function and lack the collaborative analysis of multimodal data. Furthermore, traditional warning methods generally use fixed thresholds, which are unable to adapt to individual differences and changes in condition, resulting in high false positive and false negative rates.

[0004] Given the complexity and dynamic nature of the cardiovascular system, single indicators and static thresholds are no longer sufficient to accurately predict sudden cardiac death in high-risk populations. Therefore, an intelligent prediction method that can comprehensively analyze multimodal data, dynamically assess cardiovascular coupling status, and adaptively adjust early warning strategies is urgently needed to improve the prevention and intervention of sudden cardiac death. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for intelligent analysis and prediction of multimodal data of a high-risk population for sudden cardiac death, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention, Provides a multimodal data intelligent analysis and prediction method for people at high risk of sudden cardiac death, including: Collect multimodal data of the subject under different exercise load conditions, including electrocardiogram data, blood pressure data, and exercise status data; The electrocardiogram data is projected into a three-dimensional coordinate system, the potential trajectory features are extracted, the potential trajectory is segmented, and the waveform area ratio and phase difference in each segment are calculated to obtain the spatiotemporal discrete index reflecting the degree of myocardial depolarization unevenness. Dynamic response analysis of blood pressure data was performed to extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressures, establish a blood pressure regulation response curve, perform time-series matching between the spatiotemporal dispersion index and the blood pressure regulation response curve, calculate the degree of synchronization offset, and obtain the cardiovascular coupling coefficient. The cardiovascular coupling coefficient is segmented and corrected according to the exercise status data, and a baseline reference interval is established under different activity intensities. When it is detected that the cardiovascular coupling coefficient deviates from the corresponding baseline interval for three consecutive sampling cycles, the rate of change of the spatiotemporal discrete index and the degree of abnormality of the blood pressure regulation response curve are calculated. The calculated results are compared with the preset risk level threshold. The risk warning level is determined based on the comparison results and the corresponding warning information is generated.

[0007] In an optional embodiment, The electrocardiogram data is projected into a three-dimensional coordinate system, the potential trajectory features are extracted, the potential trajectory is segmented, and the waveform area ratio and phase difference in each segment are calculated to obtain the spatiotemporal discrete index reflecting the degree of myocardial depolarization unevenness, including: Mapping the electrocardiogram data to a spatial rectangular coordinate system to obtain an initial potential trajectory, performing singular value decomposition on the initial potential trajectory, and selecting main eigenvectors according to the size of the singular values to reconstruct the noise-reduced potential trajectory; calculating motion characteristic parameters of adjacent sampling points in the denoised potential trajectory, including tangential velocity and normal acceleration, determining a dynamic segmentation threshold based on the local distribution of the motion characteristic parameters, and dividing the denoised potential trajectory into a plurality of potential trajectory segments using the dynamic segmentation threshold; Project each potential trajectory segment on three orthogonal planes, perform polar coordinate sorting based on the center of gravity of the projection point set, construct the projection contour by adding point by point, calculate the projection contour area of each potential trajectory segment on the three orthogonal planes, and take the ratio of the minimum projection contour area to the maximum projection contour area as the waveform area ratio; Calculate the potential vector angle between adjacent potential trajectory segments to obtain the phase difference feature; The weight coefficient is determined according to the energy distribution of each potential trajectory segment, and the waveform area ratio and the phase difference feature are weightedly fused to obtain a spatiotemporal discrete index reflecting the degree of uneven myocardial depolarization.

[0008] In an optional embodiment, Project each potential trajectory segment on three orthogonal planes, perform polar coordinate sorting based on the center of gravity of the projection point set, construct the projection contour by adding point by point, and calculate the projection contour area of each potential trajectory segment on the three orthogonal planes including: Project the potential trajectory segments onto three mutually orthogonal projection planes: the horizontal plane, the sagittal plane, and the coronal plane to obtain three sets of projection point sets. Calculate the barycentric coordinates of the three sets of projection point sets respectively, and translate each set of projection point sets to the corresponding barycentric coordinates. Performing cubic spline interpolation resampling on the projected point set after translation, determining the number of interpolation nodes according to the distribution density of the projected point set to obtain a uniformly distributed resampled point set, converting the resampled point set into a polar coordinate representation with the center of gravity as the origin, and sorting the points in ascending order according to the polar angle to obtain an ordered point set; Selecting the three points with the smallest polar angles in the ordered point set to construct an initial contour, calculating the area-perimeter ratio of the initial contour as a reference value, and for the points to be processed in the ordered point set, calculating the area-perimeter ratio and curvature characteristic value of the triangle formed by the points to be processed and the last two points of the current contour, and performing weighted calculation to obtain a convexity determination function value; The dynamic threshold is determined according to the convexity judgment function value of the processed points. The points to be processed whose convexity judgment function value is greater than the dynamic threshold are added to the contour point set of the corresponding plane in sequence. The contour point sets on the three planes are interpolated by tension spline to obtain the closed projected contour, and the projected contour area is calculated by the cross product of the connecting vectors of adjacent contour points.

[0009] In an optional embodiment, Perform dynamic response analysis on blood pressure data, extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressure, and establish a blood pressure regulation response curve including: Perform multi-scale wavelet decomposition on the blood pressure data, determine the optimal decomposition scale according to the inflection point of the energy distribution curve of each scale, select the decomposition coefficient corresponding to the optimal decomposition scale for signal reconstruction, and obtain a preprocessed blood pressure signal; The preprocessed blood pressure signal is extracted using an adaptive threshold to extract the systolic pressure peak point and the diastolic pressure trough point, a correction window is set based on the heart rate variability to correct abnormal points, the start and end positions of adjacent cardiac cycles are identified based on the corrected feature point sequence, the time interval from the systolic pressure peak point to the diastolic pressure trough point within the cardiac cycle is calculated, and a pressure wave conduction time difference sequence is constructed; Normalizing adjacent cardiac cycle waveforms in the preprocessed blood pressure signal and calculating the correlation coefficients between adjacent cardiac cycle waveforms; fitting the normalized cardiac cycle waveforms using a piecewise cubic spline function, calculating the mean square error and curvature variation, and constructing waveform distortion features; The mutual correlation coefficient and the waveform distortion feature are combined into a progressive feature vector, a singular value decomposition is performed on the progressive feature vector sequence, a feature rank is determined according to a singular value decay curve, and a progressive relationship matrix is established; The pressure wave conduction time difference sequence is used as the first regulation response feature, and the progressive relationship matrix is used as the second regulation response feature. The first regulation response feature and the second regulation response feature are weightedly combined, and the weighted eigenvalues are used as discrete nodes. Adjacent discrete nodes are linearly interpolated and connected to construct a blood pressure regulation response curve.

[0010] In an optional embodiment, The spatiotemporal dispersion index is time-series matched with the blood pressure regulation response curve, and the degree of synchronization offset is calculated to obtain the cardiovascular coupling coefficient, including: Construct a time series matching function that includes time delay feature terms and morphological feature terms; align the spatiotemporal discrete exponential sequence with the blood pressure regulation response curve on the time axis to obtain an initial matching point sequence; determining a path search area based on waveform change characteristics of the initial matching point sequence, extracting physiological rhythm features within the path search area, constructing a path cost function based on the physiological rhythm features, and optimizing the path cost function using a dynamic programming algorithm to obtain an optimal matching path; Calculating the time interval change sequence of adjacent matching points on the optimal matching path, extracting the fluctuation characteristics using multi-segment time domain analysis to obtain the time synchronization offset, and simultaneously extracting the amplitude difference sequence of corresponding points on the optimal matching path, constructing a nonlinear mapping function based on the distribution characteristics of the amplitude difference sequence to obtain the amplitude synchronization offset; Based on the changing trend of the time synchronization offset and the distribution range of the amplitude synchronization offset, a cardiovascular coupling feature space is established and an evaluation threshold is determined; according to the evaluation threshold, the time synchronization offset and the amplitude synchronization offset are weighted mapped to obtain a cardiovascular coupling coefficient.

[0011] In an optional embodiment, The cardiovascular coupling coefficient is calibrated in sections based on the exercise status data to establish baseline reference intervals at different activity intensities, including: Extract the peak and trough positions and amplitude information from the motion state data, calculate the time interval sequence and amplitude change sequence between adjacent peaks, determine the activity cycle characteristics based on the time interval sequence, and obtain the activity intensity characteristics based on the amplitude change sequence; construct a two-dimensional phase map of the activity cycle characteristics and activity intensity characteristics, and extract the phase trajectory characteristics; segmenting the motion state data using the phase trajectory feature to obtain data segments at different activity intensities, calculating the offset of the cardiovascular coupling coefficient in each data segment to generate a segmented correction coefficient sequence, processing the segmented correction coefficient sequence to extract the rate of change feature and the mean feature in each segment, using the rate of change feature as a real-time correction amount and the mean feature as a baseline correction amount; and performing segmented correction on the cardiovascular coupling coefficient based on the real-time correction amount and the baseline correction amount; Based on the segmented corrected cardiovascular coupling coefficient, a baseline reference interval is established under different activity intensities. When a change in activity intensity is detected, the change rate characteristics within the current data segment are used to predict the correction range of the next data segment, thereby realizing dynamic updating of the baseline reference interval.

[0012] In an optional embodiment, Calculate the rate of change of the spatiotemporal dispersion index and the degree of abnormality of the blood pressure regulation response curve, compare the calculation results with the preset risk level threshold, determine the risk warning level based on the comparison results, and generate corresponding warning information including: Perform dynamic feature decomposition on the spatiotemporal dispersion index to extract amplitude, phase, and morphological features. Calculate the oscillation intensity value based on the amplitude feature, the synchronization offset value based on the phase feature, and the distortion degree value based on the morphological feature. The combined change rate of the oscillation intensity value, synchronization offset value, and distortion degree value is determined as the change rate of the spatiotemporal dispersion index. Extracting the pressure wave conduction delay time, wave velocity change value, and waveform attenuation value from the blood pressure regulation response curve, reconstructing the pressure wave conduction delay time, wave velocity change value, and waveform attenuation value into a conduction state trajectory according to a time series, calculating the divergence value of the conduction state trajectory, determining the divergence value as the degree of abnormality in blood pressure regulation, normalizing the change rate and the degree of abnormality to obtain normalized data, performing linear regression analysis on the normalized data to obtain a slope value, determining the change trend based on the positive or negative sign of the slope value, and determining the change amplitude based on the absolute value of the slope value; Set a risk warning threshold based on the combination of change trend and change amplitude. When the change trend is positive and the change amplitude exceeds the warning threshold, extract the eigenvalue with the largest contribution in the change rate as the intervention indicator. According to the intervention indicators, early warning information including characteristic value change curve and corresponding intervention parameters is generated.

[0013] According to a second aspect of the embodiments of the present invention, Provides a multimodal data intelligent analysis and prediction system for people at high risk of sudden cardiac death, including: The first unit is used to collect multimodal data of the subject under different exercise load states, including electrocardiogram data, blood pressure data and exercise status data; The second unit is used to project the electrocardiogram data into a three-dimensional coordinate system, extract the potential trajectory features, segment the potential trajectory, calculate the waveform area ratio and phase difference in each segment, and obtain the spatiotemporal discrete index reflecting the degree of uneven myocardial depolarization; The third unit is used to perform dynamic response analysis on blood pressure data, extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressure, establish a blood pressure regulation response curve, perform time series matching between the spatiotemporal dispersion index and the blood pressure regulation response curve, calculate the degree of synchronization offset, and obtain the cardiovascular coupling coefficient; The fourth unit is used to perform segmented correction of the cardiovascular coupling coefficient based on the exercise status data, establish baseline reference intervals under different activity intensities, and calculate the rate of change of the spatiotemporal discrete index and the degree of abnormality of the blood pressure regulation response curve when it is detected that the cardiovascular coupling coefficient deviates from the corresponding baseline interval for three consecutive sampling cycles. The calculation results are compared with the preset risk level threshold, and the risk warning level is determined based on the comparison results and the corresponding warning information is generated.

[0014] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0016] In this embodiment, a multimodal data intelligent analysis and prediction method is used to effectively identify and warn people at high risk of sudden cardiac death. By projecting the electrocardiogram data into three-dimensional space and extracting the potential trajectory features, and calculating the spatiotemporal discrete index, the degree of myocardial depolarization unevenness can be accurately quantified. Compared with traditional electrocardiogram analysis methods, the accuracy of identifying abnormal patterns of cardiac electrical activity is improved. Based on the dynamic response analysis of blood pressure data and the time series matching of electrocardiogram data, a cardiovascular coupling coefficient is established, which effectively captures the abnormal state of the cardiovascular system's coordinated work, breaks through the limitations of single physiological parameter assessment, and realizes a comprehensive assessment of potential risks. The cardiovascular coupling coefficient is segmented and corrected according to the exercise state data, and an individualized baseline reference interval is established, which realizes real-time monitoring and warning at different activity intensities, greatly improving the accuracy and timeliness of the warning, reducing the false alarm rate, and providing a more reliable decision-making basis for clinical intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the process of the method for intelligent analysis and prediction of multimodal data of a high-risk population for sudden cardiac death according to an embodiment of the present invention; Figure 2 This is a histogram of parameter optimization of the convexity determination function according to an embodiment of the present invention; Figure 3 A schematic diagram of path optimization for calculating the cardiovascular coupling coefficient according to an embodiment of the present invention; Figure 4 A framework diagram for setting risk warning thresholds and generating intervention indicators for an embodiment of the present invention; Figure 5 Schematic diagram of the structure of the multimodal data intelligent analysis and prediction system for high-risk populations of sudden cardiac death according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0019] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0020] Figure 1 FIG. 1 is a flow chart of a method for intelligently analyzing and predicting multimodal data of a high-risk population for sudden cardiac death according to an embodiment of the present invention. Figure 1 As shown, the method includes: Collect multimodal data of the subject under different exercise load conditions, including electrocardiogram data, blood pressure data, and exercise status data; The electrocardiogram data is projected into a three-dimensional coordinate system, the potential trajectory features are extracted, the potential trajectory is segmented, and the waveform area ratio and phase difference in each segment are calculated to obtain the spatiotemporal discrete index reflecting the degree of myocardial depolarization unevenness. Dynamic response analysis of blood pressure data was performed to extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressures, establish a blood pressure regulation response curve, perform time-series matching between the spatiotemporal dispersion index and the blood pressure regulation response curve, calculate the degree of synchronization offset, and obtain the cardiovascular coupling coefficient. The cardiovascular coupling coefficient is segmented and corrected according to the exercise status data, and a baseline reference interval is established under different activity intensities. When it is detected that the cardiovascular coupling coefficient deviates from the corresponding baseline interval for three consecutive sampling cycles, the rate of change of the spatiotemporal discrete index and the degree of abnormality of the blood pressure regulation response curve are calculated. The calculated results are compared with the preset risk level threshold. The risk warning level is determined based on the comparison results and the corresponding warning information is generated.

[0021] In an optional embodiment, the electrocardiogram data is projected into a three-dimensional coordinate system, the potential trajectory features are extracted, the potential trajectory is segmented, and the waveform area ratio and phase difference in each segment are calculated to obtain a spatiotemporal discrete index reflecting the degree of myocardial depolarization heterogeneity, including: Mapping the electrocardiogram data to a spatial rectangular coordinate system to obtain an initial potential trajectory, performing singular value decomposition on the initial potential trajectory, and selecting main eigenvectors according to the size of the singular values to reconstruct the noise-reduced potential trajectory; calculating motion characteristic parameters of adjacent sampling points in the denoised potential trajectory, including tangential velocity and normal acceleration, determining a dynamic segmentation threshold based on the local distribution of the motion characteristic parameters, and dividing the denoised potential trajectory into a plurality of potential trajectory segments using the dynamic segmentation threshold; Project each potential trajectory segment on three orthogonal planes, perform polar coordinate sorting based on the center of gravity of the projection point set, construct the projection contour by adding point by point, calculate the projection contour area of each potential trajectory segment on the three orthogonal planes, and take the ratio of the minimum projection contour area to the maximum projection contour area as the waveform area ratio; Calculate the potential vector angle between adjacent potential trajectory segments to obtain the phase difference feature; The weight coefficient is determined according to the energy distribution of each potential trajectory segment, and the waveform area ratio and the phase difference feature are weightedly fused to obtain a spatiotemporal discrete index reflecting the degree of uneven myocardial depolarization.

[0022] For example, twelve-lead ECG data is first collected from a human. The signals from each lead are linearly combined as basis vectors in a spatial Cartesian coordinate system to construct a three-dimensional potential trajectory. The potential value at each sampling moment is mapped to a point in space, and the continuous sampling points form the initial potential trajectory. For example, for a standard cardiac cycle, a spatial trajectory containing 800 sampling points can be obtained.

[0023] When denoising the initial potential trajectory, the trajectory points are first arranged into a matrix in time sequence, and the matrix is subjected to singular value decomposition. By analyzing the energy distribution of the singular values, the eigenvectors with a cumulative contribution rate of more than 95% are selected for trajectory reconstruction. Specifically, if twenty singular values are obtained by decomposition, the eigenvectors corresponding to the first five singular values can be selected to retain the main information. After obtaining the denoised potential trajectory, the motion characteristic parameters between adjacent sampling points are calculated. The tangential velocity is obtained by the ratio of the displacement to the time interval between adjacent points, and the normal acceleration is calculated by the rate of change of the tangential velocity. These parameters are statistically analyzed using a sliding window, with the window size set to 30 sampling points. The mean and standard deviation of the parameters in the window are calculated, and the range of the mean plus or minus two times the standard deviation is used as the dynamic segmentation threshold.

[0024] The potential trajectory is segmented based on the dynamic segmentation threshold. When the motion characteristic parameter of a sampling point exceeds the threshold range, the point is marked as a segmentation point. Usually a cardiac cycle can be divided into six to eight potential trajectory segments, corresponding to each characteristic wave in the electrocardiogram. Each potential trajectory segment is projected on three orthogonal planes: xy, yz, and xz. Taking the xy plane as an example, the center of gravity of the projection point set is first calculated, and the center of gravity is used as the origin of the polar coordinates. The projection points are sorted according to the size of the polar angle. Starting from the point with the smallest polar angle, each point is connected to form a closed polygon, which is the projection contour. The triangulation method is used to calculate the area of the projection contour.

[0025] After obtaining the projected contour areas on three orthogonal planes, the maximum and minimum values are selected to calculate the ratio. For example, if the projected areas of a potential trace segment on the three planes are 100, 80, and 60 square units, respectively, then its waveform area ratio is 0.6.

[0026] When calculating the phase difference between adjacent potential trajectory segments, the line connecting the starting and ending points of each segment is used as the segment's potential vector, and the angle between adjacent vectors is calculated. Normally, the angle between adjacent segments in a normal ECG signal changes gradually; abrupt changes may indicate abnormal myocardial depolarization. When calculating the weight coefficient for each potential trajectory segment, the energy characteristics of the segment are first extracted, including trajectory length and curvature changes. These characteristics are normalized to obtain the weight coefficient. Segments with larger weight coefficients are given greater importance in the final index calculation.

[0027] Finally, the waveform area ratio and phase difference characteristics are weighted and summed. Specifically, the area ratio of each segment is multiplied by the corresponding weight coefficient and then added to the average of the phase difference characteristics of adjacent segments multiplied by the weight coefficient to obtain the local dispersion index of that segment. The weighted average of the local dispersion indices of all segments is the overall spatiotemporal dispersion index. A larger spatiotemporal dispersion index indicates a greater degree of uneven myocardial depolarization.

[0028] In practice, if the spatiotemporal dispersion index of a cardiac cycle exceeds 0.8, and the phase difference between adjacent segments shows a sudden change greater than 60 degrees, it indicates the possibility of severe myocardial depolarization abnormalities, necessitating further comprehensive evaluation in combination with other indicators. By continuously monitoring the changing trend of the spatiotemporal dispersion index, abnormal changes in myocardial function can be detected promptly.

[0029] In this embodiment, dynamic feature extraction of myocardial depolarization is achieved by projecting ECG data onto three-dimensional space and performing refined analysis. Singular value decomposition (SVD) is used for noise reduction preprocessing, effectively eliminating interference components in the signal and improving feature extraction accuracy. An adaptive segmentation method based on motion feature parameters dynamically adjusts the segmentation threshold according to signal characteristics, avoiding segmentation errors caused by fixed thresholds. By analyzing the projection features of potential trajectories on three orthogonal planes and combining waveform area ratios and phase differences, the spatial conduction characteristics of myocardial depolarization are comprehensively characterized. A weighted fusion strategy based on energy distribution emphasizes the contribution of key segments, suppresses interference from non-feature segments, and improves the characterization capability of the spatiotemporal discreteness index. This approach overcomes the limitation of traditional ECG analysis methods that focus solely on time-domain features. Through multi-dimensional feature extraction and dynamic weighted fusion, it accurately reflects the degree of heterogeneity in myocardial depolarization. This solution exhibits strong anti-interference capabilities, adaptability, and reliability, providing an important assessment basis for early warning of sudden cardiac death.

[0030] In an optional embodiment, each potential trajectory segment is projected onto three orthogonal planes, polar coordinates are sorted based on the centroid of the projection point set, and a projection contour is constructed by adding point by point. Calculating the projection contour area of each potential trajectory segment on the three orthogonal planes includes: Project the potential trajectory segments onto three mutually orthogonal projection planes: the horizontal plane, the sagittal plane, and the coronal plane to obtain three sets of projection point sets. Calculate the barycentric coordinates of the three sets of projection point sets respectively, and translate each set of projection point sets to the corresponding barycentric coordinates. Performing cubic spline interpolation resampling on the projected point set after translation, determining the number of interpolation nodes according to the distribution density of the projected point set to obtain a uniformly distributed resampled point set, converting the resampled point set into a polar coordinate representation with the center of gravity as the origin, and sorting the points in ascending order according to the polar angle to obtain an ordered point set; Selecting the three points with the smallest polar angles in the ordered point set to construct an initial contour, calculating the area-perimeter ratio of the initial contour as a reference value, and for the points to be processed in the ordered point set, calculating the area-perimeter ratio and curvature characteristic value of the triangle formed by the points to be processed and the last two points of the current contour, and performing weighted calculation to obtain a convexity determination function value; The dynamic threshold is determined according to the convexity judgment function value of the processed points. The points to be processed whose convexity judgment function value is greater than the dynamic threshold are added to the contour point set of the corresponding plane in sequence. The contour point sets on the three planes are interpolated by tension spline to obtain the closed projected contour, and the projected contour area is calculated by the cross product of the connecting vectors of adjacent contour points.

[0031] For example, the potential trajectory segment is projected on three orthogonal planes. The potential trajectory segment refers to a set of potential signal trajectory data points collected in three-dimensional space, which can be represented as a point sequence P = {(x1, y1, z1), (x2, y2, z2), ..., (x n ,y n , z n )}. Project the point sequence onto the horizontal plane (XY plane), sagittal plane (XZ plane) and coronal plane (YZ plane) respectively to obtain three sets of projection point sets P xy 、P xz and P yz For example, for P xy A set of projected points, where each point has coordinates (xᵢ, yᵢ).

[0032] Calculate the center of gravity of each projection point set, using P xy For example, the projection point set has a centroid coordinate G xy =(x g ,y g ) is calculated by averaging the x- and y-coordinates of all points. For example, for a projection point set containing 100 points, the centroid coordinates are (23.56, 42.18). The projection point set is transformed relative to the centroid so that the centroid becomes the new coordinate origin. The transformed point set is denoted as P' xy , where the coordinates of each point are (xᵢ-x g , yᵢ-y g ). xz and P yz The same operation is performed on the projected point set. Then, the translated projected point set is resampled using cubic spline interpolation. The appropriate number of interpolation nodes is determined based on the distribution characteristics of the projected point set. For example, when the original projected point set contains 100 points and is densely distributed, the number of interpolation nodes can be set to 200; when the point set is sparsely distributed, it can be set to 1.5 times the number of original points. The projected point set is resampled using the cubic spline interpolation algorithm to obtain a uniformly distributed resampled point set. xy Taking the point set as an example, after resampling, the point set P'' is obtained xy , contains 200 evenly distributed points.

[0033] Convert the resampled point set to polar coordinate representation. xy Taking a point set as an example, for a point (x, y), its polar coordinates are represented as (r, θ). It is important to note the quadrant problem when calculating angles, ensuring that θ∈[0, 2π]. For example, the polar coordinates of the point (3.2, -4.5) are represented as (5.52, 5.36). Sorting by polar angle θ from smallest to largest yields an ordered set of points, denoted as P''' xy Similarly, we get P''' xz and P'''yz Then construct the initial contour, and select the three points with the smallest polar angle from the ordered point set to form the initial contour. xy For example, if the three points with the smallest polar angle are p1, p2, and p3, the initial contour is the triangle p1p2p3. Calculate the area-perimeter ratio of the triangle as a reference value, denoted by R ref For example, when the area of the triangle is 12.5 square units and the perimeter is 15.8 units, R ref =0.79.

[0034] Process the remaining points in the ordered point set in polar angle order. For the point to be processed p, calculate the area-perimeter ratio R of the triangle formed by it and the two end points of the current contour. p At the same time, calculate the curvature eigenvalue C p , which represents the effect of adding point p on the local shape of the contour. p and C p The convexity determination function value F is obtained by weighted calculation using the weight coefficient w. p =w·R p +(1-w)·C p , where w can be set to 0.7. For example, when R p =0.85, C p =0.62, F p =0.783.

[0035] Based on the convexity judgment function value of the processed points, the dynamic threshold T is determined. Initially, T can be set to R ref 0.9 times of the original value. As the number of contour points increases, T is gradually adjusted. For example, the initial T = 0.71 is adjusted to 0.75 after processing 10 points. Only when the convexity judgment function value F of the point to be processed p is p Only when it is greater than the dynamic threshold T, p is added to the contour point set.

[0036] For P''' xy For each point in , if its convexity judgment function value is 0.92, which is greater than the current threshold 0.75, it is added to the contour point set; if it is 0.62, which is less than the threshold, it is not added. In this way, all points are processed in sequence, and finally the horizontal plane projection contour point set C is obtained. xy Similarly, we get the projection contour point set C of the sagittal and coronal planes. xz and C yz .

[0037] Perform tension spline interpolation on each projected contour point set to generate a smooth closed contour curve. The tension coefficient can be set to 0.5 to ensure that the curve is smooth while maintaining the contour characteristics. For example, C xy Contains 28 points, and generates a smooth contour curve L containing 100 points through tension spline interpolationxy Finally, calculate the projection contour area. For the horizontal plane projection contour L xy , the area can be calculated by cross product of the connecting vectors of adjacent contour points. Specifically, assuming L xy The points in the matrix are arranged in clockwise or counterclockwise order as p1, p2, ..., p n , p1, the coordinates of each point are (xᵢ, yᵢ), then the area A xy It can be obtained by calculation. For example, for a certain potential trajectory segment, the projection contour areas on three orthogonal planes are A xy =156.32 square units, A xz =142.87 square units, A yz =178.45 square units.

[0038] Through the above steps, the projected contour areas of the potential trajectory segments on three orthogonal planes are calculated. This area data can be used to analyze the morphological characteristics of the potential trajectory segments, providing an important reference for research and application in related fields.

[0039] In practical applications, parameters can be adjusted based on specific needs. For example, when processing high-noise potential signals, the number of resampling points can be increased and the convexity threshold can be adjusted to obtain smoother contours. When focusing on detailed features, the threshold can be lowered to retain more contour points. This flexible parameter adjustment allows this method to adapt to the processing needs of various potential trajectory segments.

[0040] Figure 2 This is a histogram of parameter optimization of the convexity determination function according to an embodiment of the present invention, such as Figure 2 The figure shows the results of optimizing the parameters of the convexity determination function. The horizontal axis represents the weight factor (w) ranging from 0.1 to 1.1, and the vertical axis represents the percentage of contour accuracy (70%-100%). The proposed solution (solid hollow boxes) performs well under different weights, with accuracy gradually improving as the weight factor increases: 90.2% for w = 0.1, 93.1% for w = 0.3, 94.6% for w = 0.5, 96.5% for w = 0.7, 97.4% for w = 0.9, and the optimal accuracy of 98.7% for w = 1.1. In comparison, the Ear-Clipping algorithm (dashed boxes) has significantly lower accuracy, reaching a maximum of only 83.8% (at w = 0.7), while the K-means clustering algorithm (diagonally filled boxes) performs the worst, with a maximum accuracy of only 78.3% (at w = 0.5). The proposed solution improves accuracy by 16.8% compared to the Ear-Clipping algorithm. The optimization results demonstrate the outstanding performance of this technical solution in constructing potential trajectory profiles, especially when dealing with complex morphologies and concave areas.

[0041] In this embodiment, the influence of the spatial position offset of the potential trajectory is effectively eliminated by projection on three orthogonal planes and normalization of the center of gravity. An adaptive interpolation strategy based on the point set distribution density is adopted to achieve uniform resampling of the projection points, avoiding contour distortion caused by uneven sampling. The introduction of the convexity judgment function and the dynamic threshold mechanism can intelligently identify key contour points and effectively suppress the interference of abnormal points and noise. Through the weighted fusion of the area-perimeter ratio and the curvature feature, the local geometric features of the contour are accurately grasped, and the robustness of the contour construction is improved. The tension spline interpolation method is used to construct a closed contour, which not only maintains the smoothness of the contour, but also ensures the accuracy of the key feature points. It overcomes the problem that the traditional projection contour construction method is susceptible to noise and contour distortion, significantly improves the accuracy and reliability of the projection contour area calculation, and lays a solid foundation for the subsequent analysis of myocardial depolarization heterogeneity.

[0042] In an optional embodiment, performing dynamic response analysis on blood pressure data, extracting the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressures, and establishing a blood pressure regulation response curve includes: Perform multi-scale wavelet decomposition on the blood pressure data, determine the optimal decomposition scale according to the inflection point of the energy distribution curve of each scale, select the decomposition coefficient corresponding to the optimal decomposition scale for signal reconstruction, and obtain a preprocessed blood pressure signal; The preprocessed blood pressure signal is extracted using an adaptive threshold to extract the systolic pressure peak point and the diastolic pressure trough point, a correction window is set based on the heart rate variability to correct abnormal points, the start and end positions of adjacent cardiac cycles are identified based on the corrected feature point sequence, the time interval from the systolic pressure peak point to the diastolic pressure trough point within the cardiac cycle is calculated, and a pressure wave conduction time difference sequence is constructed; Normalizing adjacent cardiac cycle waveforms in the preprocessed blood pressure signal and calculating the correlation coefficients between adjacent cardiac cycle waveforms; fitting the normalized cardiac cycle waveforms using a piecewise cubic spline function, calculating the mean square error and curvature variation, and constructing waveform distortion features; The mutual correlation coefficient and the waveform distortion feature are combined into a progressive feature vector, a singular value decomposition is performed on the progressive feature vector sequence, a feature rank is determined according to a singular value decay curve, and a progressive relationship matrix is established; The pressure wave conduction time difference sequence is used as the first regulation response feature, and the progressive relationship matrix is used as the second regulation response feature. The first regulation response feature and the second regulation response feature are weightedly combined, and the weighted eigenvalues are used as discrete nodes. Adjacent discrete nodes are linearly interpolated and connected to construct a blood pressure regulation response curve.

[0043] For example, the collected blood pressure data is subjected to multi-scale wavelet decomposition. The db4 wavelet basis function is selected to decompose the original blood pressure signal to the 8th layer. The energy distribution of the decomposition coefficients at each scale is calculated to obtain the energy percentage sequence [45.2%, 27.3%, 15.4%, 8.1%, 2.3%, 1.1%, 0.4%, 0.2%]. By calculating the first-order difference value of the energy curve [17.9%, 11.9%, 7.3%, 5.8%, 1.2%, 0.7%, 0.2%], it is found that the difference value between the 4th and 5th layers drops sharply from 5.8% to 1.2%. The 4th layer corresponding to this inflection point is determined to be the optimal decomposition scale. The wavelet coefficients corresponding to this scale are used to reconstruct the signal to obtain the preprocessed blood pressure signal, which effectively removes the baseline drift and high-frequency noise.

[0044] An adaptive thresholding method was used to extract feature points from the preprocessed blood pressure signal. The initial threshold was set at 40% of the signal amplitude, and a 120-ms sliding window was used to move across the signal. When the maximum value within the window exceeded the threshold and was a local peak, it was marked as a systolic peak. When the minimum value within the window was below the signal mean and was a local valley, it was marked as a diastolic valley. Heart rate variability was calculated for five consecutive cardiac cycles. When the variability exceeded 15%, the current window was expanded to 300 ms for outlier correction. If the interval between two adjacent systolic peaks was less than 350 ms or greater than 1200 ms, it was marked as an outlier and corrected using interpolation.

[0045] Based on the corrected sequence of characteristic points, adjacent cardiac cycles are identified. Taking the systolic peak as the starting point of the cardiac cycle, the time interval from the systolic peak to the diastolic trough within each cardiac cycle is calculated, with a typical value of 280 to 320 ms. For the time intervals of 100 consecutive cardiac cycles, a sequence of pressure wave propagation time differences is constructed, for example, [285 ms, 290 ms, 287 ms, 293 ms, 282 ms...].

[0046] To analyze waveform progression, adjacent cardiac cycle waveforms in the preprocessed blood pressure signal were normalized. Each cardiac cycle waveform was amplitude normalized to the range [0, 1]. Time normalization was also performed to uniformly define the cardiac cycle length to 1000 sampling points. The cross-correlation coefficients between adjacent normalized cardiac cycle waveforms were calculated, yielding a sequence of cross-correlation coefficients such as [0.92, 0.94, 0.89, 0.95, 0.91...].

[0047] A piecewise cubic spline function was used to fit the normalized cardiac cycle waveform. The cardiac cycle was divided into eight equal segments, with nodes set at the boundary and midpoint of each segment, for a total of 15 nodes. The mean square error (MSE) between the fitted curve and the original waveform was calculated, with typical values ranging from 0.002 to 0.015. The curvature variation of the fitted curve at characteristic points was calculated, ranging from 0.04 to 0.08 at the systolic peak and from 0.02 to 0.05 at the diastolic trough. The mean square error and curvature variation were combined to construct the waveform distortion feature vector.

[0048] The cross-correlation coefficient and waveform distortion features were combined into a progressive eigenvector, typically with a dimension of 5, containing the cross-correlation coefficient, the curvature change at the systolic peak point, the curvature change at the diastolic trough point, the mean square error along the rising edge, and the mean square error along the falling edge. Singular value decomposition was performed on the progressive eigenvector sequence for 100 consecutive cardiac cycles, yielding a singular value sequence such as [3.42, 1.87, 0.95, 0.32, 0.11]. The singular value decay curve showed a slow decay after the third singular value, indicating a characteristic rank of 3. The first three singular vectors were used to construct a progressive relationship matrix with a dimension of 5×3.

[0049] The pressure wave propagation time difference sequence was used as the first regulatory response feature, and the principal component projection value of the progressive relationship matrix was used as the second regulatory response feature. A weighted combination of the first and second regulatory response features was performed with a weight ratio of 7:3. The weighted eigenvalue sequence (e.g., [0.34, 0.37, 0.35, 0.39, 0.33...]) was used as discrete nodes, and adjacent discrete nodes were connected by linear interpolation to construct a blood pressure regulatory response curve.

[0050] In practical applications, continuous 24-hour blood pressure monitoring data from a 60-year-old male patient was analyzed at a sampling rate of 200Hz. The extracted pressure wave conduction time difference series had a mean of 295ms and a standard deviation of 18ms. The principal component explanatory power of the waveform progressive feature matrix was 82.4%. The constructed regulatory response curve showed that the patient's regulatory response value increased significantly between 6:00 and 8:00 a.m., reaching 1.45 times the daily average, indicating enhanced blood pressure regulatory sensitivity during this period, consistent with the clinical observation of an increased risk of sudden blood pressure changes in the morning.

[0051] In this embodiment, multi-scale wavelet decomposition and adaptive threshold extraction are used to precisely locate characteristic points in blood pressure signals. A correction window mechanism based on heart rate variability effectively identifies and corrects abnormal waveforms, improving the reliability of feature extraction. Normalization and cross-correlation analysis are used to accurately characterize the waveform evolution characteristics of adjacent cardiac cycles, overcoming the lack of sensitivity to waveform changes in traditional methods. Through piecewise cubic spline fitting and singular value decomposition, a quantitative representation of blood pressure waveform distortion and progressive relationships is established, effectively capturing subtle changes in blood pressure regulation. Intelligently weighted fusion of pressure wave propagation time difference and progressive relationship features is used to construct a blood pressure regulation response curve that maintains temporal continuity while reflecting the dynamic nature of waveform evolution. This method overcomes the limitations of traditional blood pressure analysis methods that focus solely on a single feature. Through multidimensional feature extraction and dynamic fusion analysis, the sensitivity of identifying abnormal blood pressure regulation is significantly improved, providing a more comprehensive and reliable basis for early warning of cardiovascular coupling abnormalities.

[0052] In an optional embodiment, the spatiotemporal dispersion index is time-series matched with the blood pressure regulation response curve, and the degree of synchronization offset is calculated to obtain the cardiovascular coupling coefficient, which includes: Construct a time series matching function that includes time delay feature terms and morphological feature terms; align the spatiotemporal discrete exponential sequence with the blood pressure regulation response curve on the time axis to obtain an initial matching point sequence; determining a path search area based on waveform change characteristics of the initial matching point sequence, extracting physiological rhythm features within the path search area, constructing a path cost function based on the physiological rhythm features, and optimizing the path cost function using a dynamic programming algorithm to obtain an optimal matching path; Calculating the time interval change sequence of adjacent matching points on the optimal matching path, extracting the fluctuation characteristics using multi-segment time domain analysis to obtain the time synchronization offset, and simultaneously extracting the amplitude difference sequence of corresponding points on the optimal matching path, constructing a nonlinear mapping function based on the distribution characteristics of the amplitude difference sequence to obtain the amplitude synchronization offset; Based on the changing trend of the time synchronization offset and the distribution range of the amplitude synchronization offset, a cardiovascular coupling feature space is established and an evaluation threshold is determined; according to the evaluation threshold, the time synchronization offset and the amplitude synchronization offset are weighted mapped to obtain a cardiovascular coupling coefficient.

[0053] The present invention provides a method for calculating the cardiovascular coupling coefficient. By performing time-series matching on the spatiotemporal discrete index and the blood pressure regulation response curve and calculating the degree of synchronization offset, a quantitative assessment of the cardiovascular coupling state can be achieved. For example, a time-series matching function including a time delay feature item and a morphological feature item is first constructed. The time delay feature item is calculated based on the offset of the signal time point, and different weights are assigned according to the size of the offset; the morphological feature item is characterized by the rate of change of the waveform curvature, and the curvature values at the corresponding points of the two curves are collected and the degree of difference is calculated. The time-series matching function can be expressed as a weighted combination of the time delay feature and the morphological feature, and the weight coefficient is obtained through sample training. For example, in clinical verification, the matching effect is best when the weight coefficients are set to 0.65 and 0.35, respectively.

[0054] The spatiotemporal discrete exponential sequence and the blood pressure regulation response curve are aligned on the time axis to obtain an initial sequence of matching points. The specific implementation process includes: first determining the starting point of the two signals, with the moment when both signals begin recording simultaneously as the zero time point; selecting the peak and trough points as feature points; and calculating the time intervals between adjacent feature points to establish an initial correspondence. For example, if a subject's physiological signals are collected for 5 minutes, the spatiotemporal discrete exponential sequence contains 23 feature points, and the blood pressure regulation response curve contains 26 feature points. After preliminary alignment, 18 pairs of initial matching points are determined.

[0055] The path search area is determined based on the waveform variation characteristics of the initial matching point sequence. The width of the search area is determined by the waveform complexity, increasing where the waveform changes dramatically and decreasing where it changes gently. In practical applications, the search width is set to 15%-25% of the matching window length. Physiological rhythm features are extracted within the determined search area, including amplitude, period, and rate of change. Amplitude describes changes in signal intensity; periodicity characterizes the signal's repetitive pattern; and rate of change reflects the instantaneous rate of change of the signal. For example, using the frequency domain index of heart rate variability as the periodic feature, the energy ratio of the low-frequency component (0.04-0.15Hz) to the high-frequency component (0.15-0.4Hz) is extracted to reflect the characteristics of autonomic nervous system regulation.

[0056] A path cost function is constructed based on the extracted physiological rhythm features. The cost function design comprehensively considers the point-to-point matching cost and path smoothness constraints. The point-to-point matching cost is calculated based on feature similarity, and the path smoothness constraint is constrained by limiting the rate of change between adjacent matching points. In practice, the rate of change between adjacent matching points is limited to ±30% to ensure the physiological plausibility of the path. A dynamic programming algorithm is used to optimize the path cost function to obtain the optimal matching path. The algorithm starts from the starting point and gradually calculates the minimum cumulative cost to each point until the globally optimal path is obtained at the end point. Experimental verification shows that this method achieves accurate matching in 98% of test samples.

[0057] The time interval variation sequence between adjacent matching points on the optimal matching path is calculated, and fluctuation characteristics are extracted using multi-segment time domain analysis. First, the entire sequence is divided into multiple time segments, each ranging from 30 to 60 seconds, dynamically adjusted based on signal stability. Statistical indicators within each time segment are calculated, including mean, standard deviation, and coefficient of variation. A trend variation model is constructed between time segments to extract fluctuation characteristics such as amplitude, frequency, and persistence. This analysis yields the time synchronization offset. For example, the average time offset for a test sample is 125 milliseconds, with a standard deviation of 37 milliseconds and a coefficient of variation of 0.296.

[0058] The amplitude difference sequence of corresponding points on the optimal matching path is simultaneously extracted, and a nonlinear mapping function is constructed based on the distribution characteristics of the amplitude difference sequence. This mapping function adopts a piecewise function form, setting different conversion coefficients for different amplitude difference ranges. Linear mapping is used for small difference ranges, and logarithmic mapping is used for large difference ranges, ensuring a reasonable distribution of output results. Through this mapping function, the amplitude synchronization offset is obtained. In clinical validation, the average amplitude synchronization offset in healthy people is 0.17, while the average value in patients with chronic diseases increases to 0.36.

[0059] Based on the changing trends of temporal synchronization offset and the distribution range of amplitude synchronization offset, a cardiovascular coupling feature space was established and evaluation thresholds were determined. The feature space is a two-dimensional plane, with the horizontal axis representing temporal synchronization offset and the vertical axis representing amplitude synchronization offset. Through large-sample data analysis, the feature space was divided into multiple regions corresponding to different coupling states. Evaluation thresholds were set based on clinical data statistics, with the temporal synchronization offset threshold set at 0.25 and the amplitude synchronization offset threshold set at 0.30.

[0060] The cardiovascular coupling coefficient is obtained by weighting the time synchronization offset and amplitude synchronization offset according to the evaluation threshold. This weighting is performed using a nonlinear weighting method, with the ratio of the time synchronization offset weight to the amplitude synchronization offset weight being 3:2. The cardiovascular coupling coefficient ranges from 0 to 1, with smaller values indicating better coupling. Clinical studies have shown that the average cardiovascular coupling coefficient in healthy individuals is 0.15±0.05, in those with mild cardiovascular dysfunction it is 0.35±0.08, in those with moderate dysfunction it is 0.55±0.10, and in those with severe dysfunction it is 0.75 or higher.

[0061] Figure 3 Schematic diagram of path optimization for calculating cardiovascular coupling coefficient according to an embodiment of the present invention. Figure 3The figure shows a detailed illustration of the path optimization process during cardiovascular coupling coefficient calculation. The horizontal axis represents the spatiotemporal discrete exponential sequence (0-600), and the vertical axis represents the blood pressure regulation response curve (0-600). The solid black line and square data points in the figure represent the optimal matching path of our proposed solution, while the dashed gray line and triangle data points represent the matching path of the traditional DTW algorithm. It can be clearly observed that at key points (such as 220, 240 and 500, 90), the offsets of our proposed solution are only 5.8% and 4.3%, respectively, while the offsets of the traditional method are significantly larger. The performance data on the right side of the figure demonstrates that the average offset rate of our proposed solution is 5.2%, while the average offset rate of the traditional DTW algorithm is as high as 16.7%, an improvement of 68.9%. The right side of the figure also shows the cardiovascular coupling indicators calculated based on the optimized path: time synchronization of 0.927, amplitude synchronization of 0.883, and the final calculated cardiovascular coupling coefficient of 0.903. These data directly demonstrate the significant advantages of our proposed solution in cardiovascular signal timing matching and coupling coefficient calculation.

[0062] Existing technologies mainly use single signal analysis or simple time alignment methods to evaluate cardiovascular coupling status, which makes it difficult to accurately reflect the dynamic correlation characteristics between signals and is easily affected by noise and baseline drift. This technical solution achieves precise alignment of the spatiotemporal discrete index and the blood pressure regulation response curve by constructing a multidimensional matching function that includes time delay and morphological characteristics, combined with a path optimization strategy based on physiological rhythm characteristics. Compared with the existing method of directly calculating signal correlation, this solution innovatively introduces an optimal matching path search mechanism based on dynamic programming, which comprehensively captures the time synchronization and amplitude synchronization characteristics between signals through multi-segment time domain analysis and nonlinear mapping. This improvement overcomes the defect of traditional methods that are insensitive to nonlinear changes in signals and improves the detection accuracy of coupling anomalies.

[0063] This improved approach takes into account the complex dynamic characteristics of cardiovascular signals. By establishing a cardiovascular coupling feature space and adaptive evaluation thresholds, it achieves precise quantification of coupling anomalies. Ultimately, it achieves significant results in the early identification of cardiovascular coupling anomalies, with higher detection sensitivity and reliability than existing technologies, providing more effective technical support for the prevention of sudden cardiac death.

[0064] In an optional embodiment, segmented correction of the cardiovascular coupling coefficient is performed based on the exercise state data to establish baseline reference intervals at different activity intensities, including: Extract the peak and trough positions and amplitude information from the motion state data, calculate the time interval sequence and amplitude change sequence between adjacent peaks, determine the activity cycle characteristics based on the time interval sequence, and obtain the activity intensity characteristics based on the amplitude change sequence; construct a two-dimensional phase map of the activity cycle characteristics and activity intensity characteristics, and extract the phase trajectory characteristics; segmenting the motion state data using the phase trajectory feature to obtain data segments at different activity intensities, calculating the offset of the cardiovascular coupling coefficient in each data segment to generate a segmented correction coefficient sequence, processing the segmented correction coefficient sequence to extract the rate of change feature and the mean feature in each segment, using the rate of change feature as a real-time correction amount and the mean feature as a baseline correction amount; and performing segmented correction on the cardiovascular coupling coefficient based on the real-time correction amount and the baseline correction amount; Based on the segmented corrected cardiovascular coupling coefficient, a baseline reference interval is established under different activity intensities. When a change in activity intensity is detected, the change rate characteristics within the current data segment are used to predict the correction range of the next data segment, thereby realizing dynamic updating of the baseline reference interval.

[0065] This embodiment provides a method for performing segmented correction of the cardiovascular coupling coefficient based on exercise state data and establishing baseline reference intervals under different activity intensities.

[0066] Specifically, motion state data is first collected, which can be obtained through sensors worn on the user's wrist, chest or other body parts. Motion state data includes but is not limited to acceleration signals, gyroscope signals, etc. At the same time, the user's electrocardiogram signal and blood pressure signal must also be collected to calculate the cardiovascular coupling coefficient. The collected motion state data is preprocessed, including filtering, denoising and standardization. For example, a 20Hz low-pass filter is applied to the acceleration data to remove high-frequency noise, and then normalization is performed to make the data range between [-1, 1]. Next, the peak and trough position and amplitude information in the motion state data are extracted. The local extreme point is identified by the sliding window method, and the window size is set to 0.5 seconds. When the value of a point is greater than all other points in the window, it is marked as a peak, and when the value of a point is less than all other points in the window, it is marked as a trough. For example, five peak points may be identified in a 5-second segment of acceleration data, located at time points 0.8 seconds, 1.9 seconds, 3.0 seconds, 4.1 seconds, and 5.2 seconds, with corresponding amplitudes of 0.85, 0.92, 0.78, 0.88, and 0.81.

[0067] Calculate the time interval sequence between adjacent peaks. For example, the time interval sequence formed by the peak points above is [1.1 seconds, 1.1 seconds, 1.1 seconds, 1.1 seconds]. Also calculate the amplitude change sequence, which is the amplitude difference between adjacent peaks, such as [0.07, -0.14, 0.10, -0.07]. The time interval sequence reflects the activity cycle, while the amplitude change sequence reflects the activity intensity.

[0068] A two-dimensional phase diagram is constructed based on the activity cycle characteristics and activity intensity characteristics. The horizontal axis represents the time interval value (unit: seconds), and the vertical axis represents the amplitude change value. The above sequence is plotted in a two-dimensional coordinate system to form a series of points. The distribution trajectory of these points is the phase trajectory feature. For example, the point set {(1.1, 0.07), (1.1, -0.14), (1.1, 0.10), (1.1, -0.07)} forms a trajectory in the phase diagram.

[0069] The motion state data is segmented by identifying clusters and change trends in the phase trajectory features. Using the density clustering algorithm DBSCAN, the distance threshold is set to 0.2 and the minimum number of points is set to 5 to divide the points in the phase map into different clusters. Each cluster represents an activity intensity state. For example, three activity intensity levels may be identified through clustering: low intensity (time interval > 1.5 seconds, amplitude change < 0.1), medium intensity (time interval 1.0-1.5 seconds, amplitude change 0.1-0.3) and high intensity (time interval < 1.0 second, amplitude change > 0.3). Corresponding to the original time series, the start and end time points of each data segment are determined. For example, the first data segment (low intensity) is 0-60 seconds, the second data segment (medium intensity) is 61-120 seconds, and the third data segment (high intensity) is 121-180 seconds.

[0070] An offset of the cardiovascular coupling coefficient is calculated within each data segment. The cardiovascular coupling coefficient can be calculated from the relationship between the R wave of the ECG signal and the corresponding blood pressure fluctuations. For each identified activity intensity segment, the difference between the cardiovascular coupling coefficient and the resting baseline value is calculated to obtain an offset sequence. For example, the offset sequence for the low-intensity segment is [0.05, 0.06, 0.04, 0.07, 0.05], the offset sequence for the medium-intensity segment is [0.15, 0.16, 0.17, 0.14, 0.15], and the offset sequence for the high-intensity segment is [0.28, 0.27, 0.29, 0.30, 0.28].

[0071] Based on the above offset sequence, a segmented correction coefficient sequence is generated. Features are extracted from the offset sequence within each data segment, including rate of change and mean features. The rate of change feature is obtained by calculating the difference between adjacent offsets. For example, the rate of change feature of the low-intensity segment is [0.01, -0.02, 0.03, -0.02]. The mean feature is the average value of the offsets in the segment, for example, the mean feature of the low-intensity segment is 0.054.

[0072] The rate of change feature is used as the real-time correction, and the mean feature is used as the baseline correction. The real-time correction is used to capture short-term fluctuations, and the baseline correction is used to reflect long-term trends. The segmented correction method for the cardiovascular coupling coefficient is: the corrected coupling coefficient = the original coupling coefficient - (baseline correction + real-time correction). For example, when the user is detected to be in a moderate-intensity activity state, the baseline correction of 0.154 for the intensity segment and the real-time correction of 0.01 at the current moment are applied to correct the original coupling coefficient of 0.85, resulting in a corrected coupling coefficient of 0.85-(0.154+0.01)=0.686.

[0073] Based on the segmented corrected cardiovascular coupling coefficient, baseline reference intervals were established for different activity intensities. By statistically analyzing the distribution characteristics of the corrected data, the mean and standard deviation were calculated, and the reference interval was set to the mean ± 1.96 times the standard deviation, covering approximately 95% of the normal value range. For example, the reference interval for low-intensity activity is [0.45-0.65], the reference interval for moderate-intensity activity is [0.60-0.80], and the reference interval for high-intensity activity is [0.75-0.95]. When the real-time monitored corrected coupling coefficient falls within the reference interval for the corresponding activity intensity, it is judged to be normal; when it exceeds the interval, there may be an abnormality.

[0074] When a change in activity intensity is detected, the system can predict the correction range for the next data segment. For example, when a user switches from low-intensity to medium-intensity activity, the system can predict the correction coefficient change trend for the initial medium-intensity segment based on the historical transition patterns between the two intensities and the rate of change characteristics of the current low-intensity segment.

[0075] Specifically, if the rate of change of the last three points in the low-intensity segment shows an upward trend, such as [0.005, 0.008, 0.012], the initial correction for the medium-intensity segment is predicted to continue increasing according to this trend, for example, to [0.018, 0.025, 0.033]. Combined with the baseline correction of 0.154 for the medium-intensity segment, the upper and lower limits of the reference interval can be dynamically updated.

[0076] Through in-depth analysis of motion state data, a two-dimensional phase image was established to characterize motion characteristics, breaking through the limitations of traditional methods that rely on a single indicator to judge motion state. The phase trajectory characteristics achieve precise segmentation of motion state, providing a reliable basis for the correction of cardiovascular coupling coefficients. A processing strategy of segmented correction coefficient sequences is adopted, and the rate of change characteristics and mean characteristics are used for real-time correction and baseline correction, respectively, achieving an adaptive response to changes at different time scales. This two-layer correction mechanism effectively eliminates the interference of motion state changes on cardiovascular coupling assessment and improves the accuracy of assessment results. By establishing a dynamically updated baseline reference interval, the system can adjust the assessment criteria in a timely manner according to changes in activity intensity, overcoming the problem that fixed reference intervals are difficult to adapt to individual differences. The adaptability and reliability of cardiovascular coupling assessment in different motion states are significantly improved, providing a more accurate judgment basis for early warning of sudden cardiac death.

[0077] In an optional embodiment, calculating the rate of change of the spatiotemporal dispersion index and the degree of abnormality of the blood pressure regulation response curve, comparing the calculation results with a preset risk level threshold, determining the risk warning level based on the comparison results, and generating corresponding warning information includes: Perform dynamic feature decomposition on the spatiotemporal dispersion index to extract amplitude, phase, and morphological features. Calculate the oscillation intensity value based on the amplitude feature, the synchronization offset value based on the phase feature, and the distortion degree value based on the morphological feature. The combined change rate of the oscillation intensity value, synchronization offset value, and distortion degree value is determined as the change rate of the spatiotemporal dispersion index. Extracting the pressure wave conduction delay time, wave velocity change value, and waveform attenuation value from the blood pressure regulation response curve, reconstructing the pressure wave conduction delay time, wave velocity change value, and waveform attenuation value into a conduction state trajectory according to a time series, calculating the divergence value of the conduction state trajectory, and determining the divergence value as the abnormality degree of blood pressure regulation; Normalizing the change rate and the degree of abnormality to obtain normalized data, performing linear regression analysis on the normalized data to obtain a slope value, determining the change trend based on the positive or negative value of the slope value, and determining the change amplitude based on the absolute value of the slope value; Set a risk warning threshold based on the combination of change trend and change amplitude. When the change trend is positive and the change amplitude exceeds the warning threshold, extract the eigenvalue with the largest contribution in the change rate as the intervention indicator. According to the intervention indicators, early warning information including characteristic value change curve and corresponding intervention parameters is generated.

[0078] This embodiment provides a blood pressure risk warning method based on the spatiotemporal dispersion index, which achieves risk warning by calculating the rate of change of the spatiotemporal dispersion index and the degree of abnormality of the blood pressure regulation response curve.

[0079] Exemplarily, the spatiotemporal discrete index is first subjected to dynamic feature decomposition. Specifically, the patient's blood pressure data is collected for 24 consecutive hours, and the data sampling frequency is 100 Hz, forming a blood pressure time series containing 86,400 sampling points. The time series is decomposed into multiple scales by wavelet transform to obtain the blood pressure fluctuation characteristics of different frequency bands. When extracting the amplitude feature, the difference between the maximum and minimum values of the waveform in each frequency band is calculated. For example, in the 0.1 Hz frequency band, the amplitude characteristic value is 15 mmHg. When extracting the phase feature, the phase difference between the waveform in each frequency band and the reference waveform is calculated. For example, in the 0.1 Hz frequency band, the phase difference is 25 degrees. When extracting the morphological feature, the kurtosis and skewness of the waveform in each frequency band are calculated. For example, in the 0.1 Hz frequency band, the kurtosis is 3.2 and the skewness is 0.5.

[0080] The oscillation intensity value is calculated based on the amplitude characteristics. The amplitude characteristics of each frequency band are weighted and summed. The weights are determined by the physiological significance of the frequency band. For example, the 0.1Hz band (reflecting sympathetic nerve activity) has a weight of 0.4, the 0.25Hz band (reflecting parasympathetic nerve activity) has a weight of 0.3, and all other frequency bands have a weight of 0.3. For one patient, the calculated oscillation intensity value was 8.7.

[0081] Calculate the synchronization offset value based on the phase characteristics. Calculate the deviation of the phase characteristics of each frequency band from the normal reference value and sum them weighted. For example, if the phase deviation of the 0.1Hz frequency band is 15 degrees and the weight is 0.4, the calculated synchronization offset value for this patient is 12.5.

[0082] The distortion level was calculated based on the morphological features. The kurtosis and skewness values of the morphological features of each frequency band were nonlinearly combined, such as by multiplying the kurtosis and skewness and taking the logarithm of the product. The distortion level for this patient was 0.85.

[0083] The combined rate of change of the oscillation intensity value, the synchronization offset value, and the distortion degree value is determined as the rate of change of the spatiotemporal dispersion index. Specifically, during continuous monitoring, the above three values are calculated once an hour and compared with the values of the previous hour to calculate the relative percentage change. At the fifth hour of monitoring, a patient had a 15% rate of change in the oscillation intensity value, an 8% rate of change in the synchronization offset value, and a 12% rate of change in the distortion degree value. The weighted combination of the three (weights 0.4, 0.3, and 0.3, respectively) yielded a rate of change of 12.2% for the spatiotemporal dispersion index.

[0084] The pressure wave conduction delay time, wave velocity change, and waveform attenuation are extracted from the blood pressure regulation response curve. By simultaneously measuring the patient's electrocardiogram and fingertip blood pressure waveforms, the time difference between the R wave peak and the corresponding pulse wave reaching the fingertip is calculated to obtain the pressure wave conduction delay time. For example, the delay time for one patient was 0.28 seconds. By continuously measuring the changes in this delay time, the wave velocity change value is calculated. For example, the wave velocity change value for this patient from a resting state to a lightly active state was 15%. By comparing the blood pressure waveform amplitude ratio at the heart and fingertips, the waveform attenuation value is calculated. For example, the attenuation value for this patient was 0.65.

[0085] The pressure wave propagation delay, wave velocity change, and waveform attenuation values are reconstructed as a conduction state trajectory in a time series. Specifically, a three-dimensional state space is constructed with these three parameters as coordinate axes, and the patient's 24-hour data forms a trajectory within this space. The Euclidean distance between this trajectory and the healthy reference trajectory is calculated to obtain the divergence value of the conduction state trajectory. The divergence value for one patient was 2.8, which was determined to represent the degree of abnormal blood pressure regulation.

[0086] Normalize the rate of change and the degree of anomaly. Using the maximum-minimum normalization method, map the rate of change and degree of anomaly to a range of 0-1. Assuming the maximum rate of change in the historical data is 30% and the minimum is 0%, the normalized value of 12.2% is 0.407. Assuming the maximum degree of anomaly is 5 and the minimum is 0, the normalized value of 2.8 is 0.56.

[0087] Linear regression analysis was performed on the normalized data. Using the normalized rate of change and degree of abnormality over 24 consecutive hours as data points, a linear regression was performed, resulting in a slope of 0.25. The trend of change was determined based on the positive or negative slope value. In this example, a slope greater than 0 indicates an increasing risk trend. The magnitude of change was determined based on the absolute value of the slope. In this example, an absolute value of 0.25 indicates a moderate magnitude of change.

[0088] Risk warning thresholds are set based on a combination of the trend and magnitude of change. The preset thresholds are: a positive slope with an absolute value greater than 0.3 indicates high risk, a positive slope with an absolute value between 0.1 and 0.3 indicates medium risk, and all other conditions indicate low risk. This patient's slope of 0.25 indicates a medium risk level.

[0089] When the trend is positive and the magnitude exceeds the warning threshold, the eigenvalue with the highest contribution to the rate of change is extracted as the intervention indicator. By analyzing the patient's oscillation intensity change rate (15%), synchronization offset change rate (8%), and distortion degree change rate (12%), it was determined that the oscillation intensity change rate had the highest contribution and was used as the intervention indicator.

[0090] Generate early warning information based on intervention indicators. This information includes: risk level (medium risk), main contributing factors (abnormal oscillation intensity values), change curve (displaying the changing trend of 24-hour oscillation intensity values), and intervention recommendations (advising patients to reduce activity intensity, avoid mood swings, and increase rest time as appropriate).

[0091] Figure 4 A framework diagram for setting risk warning thresholds and generating intervention indicators for an embodiment of the present invention is shown in FIG. Figure 4 The figure details the complete framework for setting risk warning thresholds and generating intervention indicators in this technical solution. The entire process begins with the input of the spatiotemporal dispersion index and the blood pressure regulation response curve data. It then proceeds through two parallel processing paths: the left path performs dynamic eigendecomposition on the spatiotemporal dispersion index, extracting characteristic values such as oscillation intensity (0.65) and synchronization deviation (0.42), and calculating the rate of change of the spatiotemporal dispersion index (0.82). The right path extracts parameters such as conduction delay (215ms) and waveform attenuation (0.38) from the blood pressure regulation response curve and calculates the degree of abnormal blood pressure regulation (0.76). The results from both paths are combined and normalized, followed by linear regression analysis, resulting in a slope of 0.78, indicating a positive trend and a high magnitude of change. At the decision point in the framework, the system determines whether intervention is necessary based on the conditions of "slope > 0 and magnitude > 0.7." If these conditions are met, a warning message is generated, indicating that the intervention indicator is waveform synchronization deviation. If not, intervention is determined not to be necessary. The risk warning threshold setting table on the left clearly defines the amplitude thresholds corresponding to different risk levels: low risk is 0.3-0.5, and high risk is >0.7. The key intervention indicator table on the right shows the characteristic values that contribute most to risk warning: synchronization offset (contribution 0.42) and wave velocity change (contribution 0.35). This framework diagram shows how this technical solution can achieve objective assessment of blood pressure regulation risks and generate precise intervention indicators through multi-feature collaborative analysis, linear regression and threshold determination. Compared with traditional single threshold comparison or outlier detection methods, it can more comprehensively assess risk status and provide targeted intervention recommendations.

[0092] Existing technologies primarily rely on single indicator thresholds or simple trend analysis for risk warnings, which fail to accurately reflect the complex dynamic changes of the cardiovascular system and lack targeted intervention guidance. This technical solution extracts the multidimensional features of spatiotemporal discrete indices through dynamic feature decomposition, combined with analysis of the conduction state trajectory of the blood pressure regulation response curve, to establish a comprehensive risk assessment system.

[0093] Compared to existing technologies, this solution innovatively incorporates a combined rate-of-change analysis of oscillation intensity, synchronization offset, and distortion. Simultaneously, by reconstructing the state trajectory of pressure wave transmission characteristics, it accurately characterizes the anomaly's development process. This improvement overcomes the limitations of traditional methods, which often struggle to identify subtle changes in the early stages, and improves the timeliness of early warnings.

[0094] This improved approach is based on the progressive nature of cardiovascular risk evolution. By establishing a multi-level early warning mechanism based on the trend and magnitude of change, a precise stratification of risk levels is achieved. By extracting the most contributing eigenvalues as intervention indicators, this provides clear guidance for clinical intervention. Ultimately, this approach significantly improves the prevention of sudden cardiac death, achieving higher early warning accuracy and more practical intervention guidance than existing technologies.

[0095] Figure 5 FIG. 1 is a schematic diagram of the structure of a multimodal data intelligent analysis and prediction system for a high-risk population for sudden cardiac death according to an embodiment of the present invention. Figure 5 As shown, the system includes: The first unit is used to collect multimodal data of the subject under different exercise load states, including electrocardiogram data, blood pressure data and exercise status data; The second unit is used to project the electrocardiogram data into a three-dimensional coordinate system, extract the potential trajectory features, segment the potential trajectory, calculate the waveform area ratio and phase difference in each segment, and obtain the spatiotemporal discrete index reflecting the degree of uneven myocardial depolarization; The third unit is used to perform dynamic response analysis on blood pressure data, extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressure, establish a blood pressure regulation response curve, perform time series matching between the spatiotemporal dispersion index and the blood pressure regulation response curve, calculate the degree of synchronization offset, and obtain the cardiovascular coupling coefficient; The fourth unit is used to perform segmented correction of the cardiovascular coupling coefficient based on the exercise status data, establish baseline reference intervals under different activity intensities, and calculate the rate of change of the spatiotemporal discrete index and the degree of abnormality of the blood pressure regulation response curve when it is detected that the cardiovascular coupling coefficient deviates from the corresponding baseline interval for three consecutive sampling cycles. The calculation results are compared with the preset risk level threshold, and the risk warning level is determined based on the comparison results and the corresponding warning information is generated.

[0096] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0097] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0098] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent analysis and prediction of multimodal data of a population at high risk of sudden cardiac death, characterized in that: include: Collect multimodal data of the subject under different exercise load conditions, including electrocardiogram data, blood pressure data, and exercise status data; The electrocardiogram data is projected into a three-dimensional coordinate system, the potential trajectory features are extracted, the potential trajectory is segmented, and the waveform area ratio and phase difference in each segment are calculated to obtain the spatiotemporal discrete index reflecting the degree of myocardial depolarization unevenness. Dynamic response analysis of blood pressure data was performed to extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressures, establish a blood pressure regulation response curve, perform time-series matching between the spatiotemporal dispersion index and the blood pressure regulation response curve, calculate the degree of synchronization offset, and obtain the cardiovascular coupling coefficient. The cardiovascular coupling coefficient is segmented and corrected according to the exercise status data, and a baseline reference interval is established under different activity intensities. When it is detected that the cardiovascular coupling coefficient deviates from the corresponding baseline interval for three consecutive sampling cycles, the rate of change of the spatiotemporal discrete index and the degree of abnormality of the blood pressure regulation response curve are calculated. The calculated results are compared with the preset risk level threshold. The risk warning level is determined based on the comparison results and the corresponding warning information is generated.

2. The method according to claim 1, characterized in that The electrocardiogram data is projected into a three-dimensional coordinate system, the potential trajectory features are extracted, the potential trajectory is segmented, and the waveform area ratio and phase difference in each segment are calculated to obtain the spatiotemporal discrete index reflecting the degree of myocardial depolarization unevenness, including: Mapping the electrocardiogram data to a spatial rectangular coordinate system to obtain an initial potential trajectory, performing singular value decomposition on the initial potential trajectory, and selecting main eigenvectors according to the size of the singular values to reconstruct the noise-reduced potential trajectory; calculating motion characteristic parameters of adjacent sampling points in the denoised potential trajectory, including tangential velocity and normal acceleration, determining a dynamic segmentation threshold based on the local distribution of the motion characteristic parameters, and dividing the denoised potential trajectory into a plurality of potential trajectory segments using the dynamic segmentation threshold; Project each potential trajectory segment on three orthogonal planes, perform polar coordinate sorting based on the center of gravity of the projection point set, construct the projection contour by adding point by point, calculate the projection contour area of each potential trajectory segment on the three orthogonal planes, and take the ratio of the minimum projection contour area to the maximum projection contour area as the waveform area ratio; Calculate the potential vector angle between adjacent potential trajectory segments to obtain the phase difference feature; The weight coefficient is determined according to the energy distribution of each potential trajectory segment, and the waveform area ratio and the phase difference feature are weightedly fused to obtain a spatiotemporal discrete index reflecting the degree of uneven myocardial depolarization.

3. The method according to claim 2, characterized in that Project each potential trajectory segment on three orthogonal planes, perform polar coordinate sorting based on the center of gravity of the projection point set, construct the projection contour by adding point by point, and calculate the projection contour area of each potential trajectory segment on the three orthogonal planes including: Project the potential trajectory segments onto three mutually orthogonal projection planes: the horizontal plane, the sagittal plane, and the coronal plane to obtain three sets of projection point sets. Calculate the barycentric coordinates of the three sets of projection point sets respectively, and translate each set of projection point sets to the corresponding barycentric coordinates. Performing cubic spline interpolation resampling on the projected point set after translation, determining the number of interpolation nodes according to the distribution density of the projected point set to obtain a uniformly distributed resampled point set, converting the resampled point set into a polar coordinate representation with the center of gravity as the origin, and sorting the points in ascending order according to the polar angle to obtain an ordered point set; Selecting the three points with the smallest polar angles in the ordered point set to construct an initial contour, calculating the area-perimeter ratio of the initial contour as a reference value, and for the points to be processed in the ordered point set, calculating the area-perimeter ratio and curvature characteristic value of the triangle formed by the points to be processed and the last two points of the current contour, and performing weighted calculation to obtain a convexity determination function value; The dynamic threshold is determined according to the convexity judgment function value of the processed points. The points to be processed whose convexity judgment function value is greater than the dynamic threshold are added to the contour point set of the corresponding plane in sequence. The contour point sets on the three planes are interpolated by tension spline to obtain the closed projected contour, and the projected contour area is calculated by the cross product of the connecting vectors of adjacent contour points.

4. The method according to claim 1, wherein Perform dynamic response analysis on blood pressure data, extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressure, and establish a blood pressure regulation response curve including: Perform multi-scale wavelet decomposition on the blood pressure data, determine the optimal decomposition scale according to the inflection point of the energy distribution curve of each scale, select the decomposition coefficient corresponding to the optimal decomposition scale for signal reconstruction, and obtain a preprocessed blood pressure signal; The preprocessed blood pressure signal is extracted using an adaptive threshold to extract the systolic pressure peak point and the diastolic pressure trough point, a correction window is set based on the heart rate variability to correct abnormal points, the start and end positions of adjacent cardiac cycles are identified based on the corrected feature point sequence, the time interval from the systolic pressure peak point to the diastolic pressure trough point within the cardiac cycle is calculated, and a pressure wave conduction time difference sequence is constructed; Normalizing adjacent cardiac cycle waveforms in the preprocessed blood pressure signal and calculating the correlation coefficients between adjacent cardiac cycle waveforms; fitting the normalized cardiac cycle waveforms using a piecewise cubic spline function, calculating the mean square error and curvature variation, and constructing waveform distortion features; The mutual correlation coefficient and the waveform distortion feature are combined into a progressive feature vector, a singular value decomposition is performed on the progressive feature vector sequence, a feature rank is determined according to a singular value decay curve, and a progressive relationship matrix is established; The pressure wave conduction time difference sequence is used as the first regulation response feature, and the progressive relationship matrix is used as the second regulation response feature. The first regulation response feature and the second regulation response feature are weightedly combined, and the weighted eigenvalues are used as discrete nodes. Adjacent discrete nodes are linearly interpolated and connected to construct a blood pressure regulation response curve.

5. The method according to claim 1, wherein The spatiotemporal dispersion index is time-series matched with the blood pressure regulation response curve, and the degree of synchronization offset is calculated to obtain the cardiovascular coupling coefficient, including: Construct a time series matching function that includes time delay feature terms and morphological feature terms; align the spatiotemporal discrete exponential sequence with the blood pressure regulation response curve on the time axis to obtain an initial matching point sequence; determining a path search area based on waveform change characteristics of the initial matching point sequence, extracting physiological rhythm features within the path search area, constructing a path cost function based on the physiological rhythm features, and optimizing the path cost function using a dynamic programming algorithm to obtain an optimal matching path; Calculating the time interval change sequence of adjacent matching points on the optimal matching path, extracting the fluctuation characteristics using multi-segment time domain analysis to obtain the time synchronization offset, and simultaneously extracting the amplitude difference sequence of corresponding points on the optimal matching path, constructing a nonlinear mapping function based on the distribution characteristics of the amplitude difference sequence to obtain the amplitude synchronization offset; Based on the changing trend of the time synchronization offset and the distribution range of the amplitude synchronization offset, a cardiovascular coupling feature space is established and an evaluation threshold is determined; according to the evaluation threshold, the time synchronization offset and the amplitude synchronization offset are weighted mapped to obtain a cardiovascular coupling coefficient.

6. The method according to claim 1, characterized in that The cardiovascular coupling coefficient is calibrated in sections based on the exercise status data to establish baseline reference intervals at different activity intensities, including: Extract the peak and trough positions and amplitude information from the motion state data, calculate the time interval sequence and amplitude change sequence between adjacent peaks, determine the activity cycle characteristics based on the time interval sequence, and obtain the activity intensity characteristics based on the amplitude change sequence; construct a two-dimensional phase map of the activity cycle characteristics and activity intensity characteristics, and extract the phase trajectory characteristics; segmenting the motion state data using the phase trajectory feature to obtain data segments at different activity intensities, calculating the offset of the cardiovascular coupling coefficient in each data segment to generate a segmented correction coefficient sequence, processing the segmented correction coefficient sequence to extract the rate of change feature and the mean feature in each segment, using the rate of change feature as a real-time correction amount and the mean feature as a baseline correction amount; and performing segmented correction on the cardiovascular coupling coefficient based on the real-time correction amount and the baseline correction amount; Based on the segmented corrected cardiovascular coupling coefficient, a baseline reference interval is established under different activity intensities. When a change in activity intensity is detected, the change rate characteristics within the current data segment are used to predict the correction range of the next data segment, thereby realizing dynamic updating of the baseline reference interval.

7. The method according to claim 1, characterized in that Calculate the rate of change of the spatiotemporal dispersion index and the degree of abnormality of the blood pressure regulation response curve, compare the calculation results with the preset risk level threshold, determine the risk warning level based on the comparison results, and generate corresponding warning information including: Perform dynamic feature decomposition on the spatiotemporal dispersion index to extract amplitude, phase, and morphological features. Calculate the oscillation intensity value based on the amplitude feature, the synchronization offset value based on the phase feature, and the distortion degree value based on the morphological feature. The combined change rate of the oscillation intensity value, synchronization offset value, and distortion degree value is determined as the change rate of the spatiotemporal dispersion index. Extracting the pressure wave conduction delay time, wave velocity change value, and waveform attenuation value from the blood pressure regulation response curve, reconstructing the pressure wave conduction delay time, wave velocity change value, and waveform attenuation value into a conduction state trajectory according to a time series, calculating the divergence value of the conduction state trajectory, determining the divergence value as the degree of abnormality in blood pressure regulation, normalizing the change rate and the degree of abnormality to obtain normalized data, performing linear regression analysis on the normalized data to obtain a slope value, determining the change trend based on the positive or negative sign of the slope value, and determining the change amplitude based on the absolute value of the slope value; Set a risk warning threshold based on the combination of change trend and change amplitude. When the change trend is positive and the change amplitude exceeds the warning threshold, extract the eigenvalue with the largest contribution in the change rate as the intervention indicator. According to the intervention indicators, early warning information including characteristic value change curve and corresponding intervention parameters is generated.

8. A multimodal data intelligent analysis and prediction system for people at high risk of sudden cardiac death, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect multimodal data of the subject under different exercise load states, including electrocardiogram data, blood pressure data and exercise status data; The second unit is used to project the electrocardiogram data into a three-dimensional coordinate system, extract the potential trajectory features, segment the potential trajectory, calculate the waveform area ratio and phase difference in each segment, and obtain the spatiotemporal discrete index reflecting the degree of uneven myocardial depolarization; The third unit is used to perform dynamic response analysis on blood pressure data, extract the pressure wave transmission time difference and waveform progression relationship between systolic and diastolic pressure, establish a blood pressure regulation response curve, perform time series matching between the spatiotemporal dispersion index and the blood pressure regulation response curve, calculate the degree of synchronization offset, and obtain the cardiovascular coupling coefficient; The fourth unit is used to perform segmented correction of the cardiovascular coupling coefficient based on the exercise status data, establish baseline reference intervals under different activity intensities, and calculate the rate of change of the spatiotemporal discrete index and the degree of abnormality of the blood pressure regulation response curve when it is detected that the cardiovascular coupling coefficient deviates from the corresponding baseline interval for three consecutive sampling cycles. The calculation results are compared with the preset risk level threshold, and the risk warning level is determined based on the comparison results and the corresponding warning information is generated.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images