Cardiovascular early warning method and system based on millimeter wave radar and physiological signal decoupling

Abstract

The invention relates to the field of cardiovascular disease monitoring and early warning, and provides a cardiovascular early warning method and system based on millimeter wave radar and physiological signal decoupling, and the method comprises the steps: obtaining a regional synchronization signal set through time-space domain secondary mapping according to an original radar data stream, mapping the regional synchronization signal set into a modal component set, inputting the modal component set into a deep residual network to generate a physiological feature vector, obtaining a body fluid retention index based on the regional synchronization signal set; generating a linkage feature matrix according to the physiological feature vector and the body fluid retention index, and generating an interaction enhancement coefficient through danger mode recognition; according to the physiological feature vector and the linkage feature matrix, generating a grading early warning decision instruction by combining an interaction enhancement coefficient through dual-path risk assessment; and generating a baseline offset vector based on the comprehensive risk evaluation value, and outputting an adaptive decision parameter set in combination with a clinical feedback signal. According to the method, multi-region physiological signal decoupling and cross-disease pathology association logic are fused, and a multi-mode intelligent early warning and parameter adaptive closed-loop calibration mechanism of cardiovascular events is constructed.

Description

Technical Field

[0001] This invention relates to the field of cardiovascular disease monitoring and early warning, and in particular to a cardiovascular early warning method and system based on the decoupling of millimeter-wave radar and physiological signals. Background Technology

[0002] With the deepening application of millimeter-wave radar sensing technology in the field of smart healthcare, the synchronous reconstruction of cross-regional physiological characteristics and the collaborative early warning of multi-dimensional risks have become key technologies for ensuring cardiovascular health and safety. How to effectively separate weak physiological signals from multiple sources such as the chest, abdomen, and limbs in complex, multi-interference monitoring scenarios, and to construct a correlation between the risk of heart failure and cerebral infarction, avoiding the problem of missed warnings due to treating different diseases as isolated events, has become a key technical challenge for non-contact monitoring systems in the process of moving from single-feature detection to system-level pathological coupling and dynamic adaptive calibration.

[0003] Chinese patent application CN120611221B discloses a method for classifying and identifying vascular data based on a large language model and clustering algorithm. The method includes: using a large language model to retrieve N literature on vascular lesions; extracting features and data related to vascular lesions from each literature as a single literature record; selecting A distinct features from the N records, discarding A1 features with low occurrence frequencies; filtering out a single literature record from the N records that contains all A2 features or is missing only one of the A2 features, thus selecting N1 records; filling in the missing feature data using the remaining literature data; calculating the weight value of each feature based on the occurrence frequency of the A2 features in the N records; obtaining data for each of the A2 features for each blood vessel in M ​​clinical patients, recording M clinical data, and actively labeling the M1 clinical data; clustering using the labeled clinical data and the weight values ​​of each feature; inputting the unlabeled clinical data and the N1 literature records into a clustering model to obtain and label each clustering result.

[0004] However, current technology still faces many challenges. In non-contact monitoring scenarios at home, the monitored subject's position is frequently changing and often obstructed by bedding. Traditional single-point radar-based monitoring technology struggles to decouple and synchronously extract physiological signals from the chest, abdomen, and limbs in a highly interfering environment. More critically, existing early warning logic often treats heart failure and cerebral infarction as isolated pathological events, ignoring the intrinsic pathological link between atrial fibrillation, which is both a trigger for heart failure and a source of emboli in cerebral infarction. When patients are in the early stages of a synergistic deterioration of cardiovascular and cerebrovascular risks, traditional discrete monitoring and early warning strategies struggle to identify latent pathological associations between cross-regional indicators because single indicators of fluid retention or arrhythmia have not yet reached preset independent alarm thresholds. This lack of quantification regarding latent synergistic risks can lead to severe missed warnings during critical windows of nonlinear risk enhancement, preventing healthcare workers or family members from intercepting early-stage cardioembolic events or acute circulatory decompensation, thus causing the monitored subject to miss the golden intervention time and endangering their life. Summary of the Invention

[0005] To achieve the above objectives, this invention provides a cardiovascular early warning method based on the decoupling of millimeter-wave radar and physiological signals, the specific technical solution of which is as follows:

[0006] Based on the collected raw radar data stream, a focused echo stream is obtained, and based on the spatiotemporal quadratic mapping function, the focused echo stream is transformed into a regional synchronization signal set containing the thoracic cavity channel, abdominal channel, and peripheral limb channel.

[0007] The regional synchronization signal set is mapped to a modal component set and input into a deep residual network to generate a physiological feature vector containing beat-by-beat heart rate interval, atrial fibrillation probability score and pulse wave conduction time. The fluid retention index is obtained based on the abdominal and chest phase data sequences of the regional synchronization signal set.

[0008] Based on the physiological feature vector and fluid retention index, the heart failure-cerebral infarction correlation mapping logic is executed to generate a linkage feature matrix, and the risk pattern recognition logic is executed based on the linkage feature matrix to generate the interaction enhancement coefficient;

[0009] The dual-path risk assessment logic is executed based on physiological feature vectors and linkage feature matrices to generate independent risk score pairs. The digital fusion logic is then combined with the interaction enhancement coefficient to obtain a comprehensive risk assessment value. Finally, the comprehensive risk assessment value is mapped to a graded early warning decision instruction using segmented threshold discrimination logic.

[0010] Based on the comprehensive risk assessment value, a baseline offset vector is generated. Combined with the clinical feedback signal obtained from the medical interactive terminal, incremental adaptive calibration logic is executed on the initial decision parameter set, which includes the fusion compensation coefficient and the early warning boundary threshold, and an adaptive decision parameter set is output.

[0011] Furthermore, the method for generating the regional synchronization signal set includes:

[0012] An adaptive spatial beamforming operator based on spatial topology compensation is constructed to perform complex weighted summation on the acquired raw radar data stream to generate a focused echo stream for a specific anatomical region;

[0013] A logical judgment matrix is ​​constructed based on the collected target distance and azimuth angle. The signal components falling within the corresponding spatial thresholds of the chest, abdomen and limbs are extracted from the focused echo stream through a spatiotemporal quadratic mapping function. The signals are then reconstructed to obtain a regional synchronous signal set composed of the chest cavity channel, the abdominal channel and the limb peripheral channel.

[0014] Furthermore, the method for obtaining the physiological feature vector includes:

[0015] The regional synchronization signal set is mapped to the sum of a finite number of narrowband mode functions using variational mode decomposition operators. The global minimum is then performed to obtain the modal component set, with the goal of minimizing the sum of the estimated bandwidths of each narrowband mode function.

[0016] A deep residual network is used to map the modal components of the chest cavity to the beat-by-beat heart rate interval, and the modal components of the chest cavity, abdomen and peripheral limbs to the atrial fibrillation probability score. The spatiotemporal cross-correlation peak time delay of the chest cavity channel and the peripheral limb channel is mapped to the pulse wave conduction time, generating a physiological feature vector.

[0017] Furthermore, the method for generating the fluid retention index includes: during the monitoring period, retrieving the abdominal phase data sequence of the abdominal channel and the chest phase data sequence of the thoracic channel, performing a time integral operation on the absolute value of the difference between the abdominal phase data sequence and the chest phase data sequence, and generating the fluid retention index by combining a preset attenuation compensation factor and a static reflection reference intensity.

[0018] Furthermore, the method for generating the interaction enhancement coefficient includes:

[0019] Based on the physiological feature vector and fluid retention index, the heart failure-cerebral infarction correlation mapping logic is executed to generate a linkage feature matrix containing covariance components, vascular resistance evolution components, arrhythmia components, and nonlinear enhancement components.

[0020] Based on the linkage feature matrix, the dangerous pattern recognition logic is executed. The risk probability mapping value is generated by combining the normalized exponential function. The risk probability mapping value is multiplied by a nonlinear enhancement exponent raised to the power of a preset risk coupling constant to obtain the interaction enhancement coefficient.

[0021] Furthermore, the execution method of the heart failure-cerebral infarction correlation mapping logic includes:

[0022] Construct a linkage feature matrix, and perform component mapping on the linkage feature matrix using physiological feature vectors and fluid retention index. Specifically, this includes generating covariance components using the covariance operators of atrial fibrillation probability score and fluid retention index within the monitoring period, and using the covariance components as the first row and first column elements of the linkage feature matrix.

[0023] Calculate the first derivative of the pulse wave conduction time with respect to the sampling time, and use the product of the first derivative and the preset first weighting coefficient as the vascular resistance evolution component. Map the vascular resistance evolution component to the first row and second column element of the linkage feature matrix.

[0024] The heart rate interval variation of the heart rate interval under adjacent sampling periods is obtained, and the product of the reciprocal of the heart rate interval variation and the preset second weighting coefficient is used as the arrhythmia component. The arrhythmia component is mapped to the second row and first column element of the linkage feature matrix.

[0025] The implicit synergistic features between the physiological feature vector and the fluid retention index are extracted using the residual correlation operator. The product of the implicit synergistic features and the preset third weighting coefficient is used as a nonlinear enhancement component. The nonlinear enhancement component is mapped to the second row and second column element of the linkage feature matrix. The values ​​of the preset first weighting coefficient, the preset second weighting coefficient, and the preset third weighting coefficient are preset values ​​based on the basic medical history record of the target individual.

[0026] Furthermore, the mapping method for the hierarchical early warning decision instructions includes:

[0027] Based on the linkage feature matrix and physiological feature vector, a dual-path risk assessment logic is executed to generate independent risk score pairs containing heart failure risk index and cerebral infarction risk index;

[0028] A digital fusion logic is performed on the independent risk score pairs and the interaction enhancement coefficient. The product of the heart failure risk index and the preset first fusion compensation coefficient, the product of the cerebral infarction risk index and the preset second fusion compensation coefficient, and the product of the interaction enhancement coefficient and the preset third fusion compensation coefficient are summed to generate a comprehensive risk assessment value.

[0029] Based on the comprehensive risk assessment value, the segmented threshold discrimination logic is executed. Combined with the preset early warning boundary threshold, the risk judgment interval where the comprehensive risk assessment value is located is determined, and the hierarchical early warning decision instruction with the status bit assigned to the first-level flag bit, the second-level flag bit, or the third-level flag bit is output.

[0030] Furthermore, the execution method of the dual-path risk assessment logic includes:

[0031] The covariance component and nonlinear enhancement component in the linkage feature matrix are used as dynamic calibration factors to perform real-time weight correction on each feature component in the physiological feature vector.

[0032] The dual-path risk assessment logic includes a synchronous heart failure risk assessment path and a stroke risk assessment path. In the heart failure risk assessment path, a first set of features consisting of beat-by-beat heart rate interval, fluid retention index, and heart rate variability is transformed by a first nonlinear feature mapping operator and then aggregated and summed with the first feature weight coefficients corrected by the dynamic calibration factor to generate a heart failure risk index. The value of heart rate variability is obtained based on the statistical analysis of the standard deviation of the beat-by-beat heart rate interval within a preset monitoring window.

[0033] In the stroke risk assessment pathway, the second feature set, consisting of atrial fibrillation probability score, pulse wave transit time, and blood pressure variability, is transformed by the second nonlinear feature mapping operator and then aggregated and summed with the second feature weight coefficients corrected by the dynamic calibration factor to generate a stroke risk index; the value of blood pressure variability is calculated based on the fluctuation dispersion of the pulse wave transit time within a preset time window.

[0034] Furthermore, the method for outputting the adaptive decision parameter set includes:

[0035] Based on the timestamp index of the sampling time, the historical data stream of the accumulated comprehensive risk assessment values ​​from multiple sampling periods is sorted into a historical time series. The static physiological baseline of the historical time series within the preset resting period is extracted, and a baseline offset vector is generated based on the instantaneous comprehensive risk assessment value and the static physiological baseline.

[0036] The risk judgment deviation value is calculated based on the comprehensive risk assessment value and the clinical feedback signal obtained from the medical interactive terminal. The partial derivative operation is performed on the baseline offset vector to obtain the feature offset gradient operator. Combined with the preset learning rate operator, the initial decision parameter set containing the first, second and third fusion compensation coefficients and the warning boundary threshold is incrementally adaptively calibrated, and the adaptive decision parameter set is output.

[0037] A cardiovascular early warning system based on millimeter-wave radar and physiological signal decoupling is used to implement the above-mentioned cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling. The system includes a regional signal decoupling module, a physiological feature extraction module, a risk linkage module, an intelligent early warning decision module, and an adaptive calibration module.

[0038] The regional signal decoupling module is used to obtain a focused echo stream based on the acquired raw radar data stream, and to convert the focused echo stream into a regional synchronization signal set containing the thoracic cavity channel, abdominal channel and peripheral limb channel according to the spatiotemporal domain quadratic mapping function.

[0039] The physiological feature extraction module is used to map the regional synchronization signal set into a modal component set and input it into a deep residual network to generate a physiological feature vector containing beat-by-beat heart rate interval, atrial fibrillation probability score and pulse wave conduction time, and to obtain the fluid retention index based on the abdominal and chest phase data sequences of the regional synchronization signal set.

[0040] The risk linkage module is used to generate a linkage feature matrix by performing heart failure-cerebral infarction correlation mapping logic based on physiological feature vectors and fluid retention index, and to perform hazard pattern recognition logic based on the linkage feature matrix to generate interaction enhancement coefficients.

[0041] The intelligent early warning decision module is used to execute dual-path risk assessment logic based on physiological feature vectors and linkage feature matrices to generate independent risk score pairs, combine interaction enhancement coefficients to execute digital fusion logic to obtain a comprehensive risk evaluation value, and use segmented threshold discrimination logic to map the comprehensive risk evaluation value into a graded early warning decision instruction.

[0042] The adaptive calibration module generates a baseline offset vector based on the comprehensive risk assessment value, combines it with the clinical feedback signal obtained from the medical interactive terminal, performs incremental adaptive calibration logic on the initial decision parameter set including the fusion compensation coefficient and the early warning boundary threshold, and outputs an adaptive decision parameter set.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention constructs an adaptive spatial beamforming operator based on spatial topology compensation to map the original radar echo stream into a virtual probe with spatial directivity, thereby achieving directional focusing of electromagnetic energy at specific anatomical locations on the human body. This solves the problem of signal masking in complex and multi-interference monitoring scenarios and completes the physical location decoupling and digital synchronous mapping of weak physiological signals in the chest, abdomen, and limb regions from the data source.

[0045] This invention establishes a jump connection architecture of variational mode decomposition operator and deep residual network to map mixed regional synchronization signals into orthogonal modal components that characterize physiological rhythms, thereby achieving orthogonal separation of myocardial micro-displacement features and random body motion interference. This solves the problem of difficulty in digitally representing weak physiological motion information caused by limb micro-swaying in complex multi-interference monitoring scenarios.

[0046] This invention constructs a linkage feature matrix containing covariance components and nonlinear enhancement components, and uses hazard pattern recognition logic to map the correlation between cardiac power output and peripheral circulation feedback into interactive enhancement coefficients. This enables nonlinear quantitative enhancement of implicit synergistic risks that are difficult to characterize with single parameters, and solves the problem of missed warnings in traditional monitoring logic when a single disease indicator is within a preset safe range but the synergistic state is abnormal due to isolated assessment of heart failure and cerebral infarction risks.

[0047] This invention constructs a decision-making system consisting of dual-path risk assessment logic and digital fusion logic. It utilizes a linkage feature matrix and interaction enhancement coefficient to perform collaborative quantification and nonlinear compensation on discrete risk components, thereby realizing the programmatic transformation from multidimensional heterogeneous physiological characteristic assessment to deterministic graded early warning instructions. This solves the problem of missed early warnings in traditional monitoring logic caused by the failure of a single disease indicator to reach the alarm threshold but the risk path collaborative deterioration.

[0048] This invention constructs a historical time series that reflects the evolution trajectory of individual risk, and uses baseline offset vectors and clinical feedback signals to perform incremental adaptive calibration on the initial decision parameter set. This achieves closed-loop alignment between the early warning decision logic and the dynamic physiological characteristics of the monitored individual, solving the problem of early warning threshold failure caused by individual physiological benchmark differences and long-term state drift in traditional monitoring systems. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the principle of the cardiovascular early warning method based on the decoupling of millimeter-wave radar and physiological signals according to the present invention.

[0051] Figure 2 This is a schematic diagram illustrating the principle of the present invention, from the MIMO antenna array to the non-anatomical region of the human body;

[0052] Figure 3 This is a schematic diagram illustrating the principle of physiological feature vector extraction based on monitoring different anatomical regions of the human body according to the present invention;

[0053] Figure 4 This is a functional block diagram of the cardiovascular early warning system based on millimeter-wave radar and physiological signal decoupling of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Example 1:

[0056] Please see Figure 1 As shown, this embodiment provides a cardiovascular early warning method based on the decoupling of millimeter-wave radar and physiological signals, including:

[0057] Step S1000, based on the collected raw radar data stream Obtain focused echo flow And based on the spatiotemporal quadratic mapping function, the focused echo flow is... Transformed into a cavity containing a thoracic passage Abdominal passage and peripheral channels Regional synchronization signal set .

[0058] Specifically, this step aims to use the echo characteristics of individual targets in complex detection environments as the data processing object, utilizing the acquired raw radar data stream. By using an adaptive spatial beamforming operator, electromagnetic energy in physical space is focused onto different anatomical locations of the human body, mapping the chaotic and superimposed physical reflected waves into a focused echo stream with spatial isolation. Furthermore, a distance-gating mapping function is used to transform the focused echo stream into one that includes the thoracic cavity passage. Abdominal passage and peripheral channels Regional synchronization signal set At the data source, the physical location decoupling and digital logical mapping of micro-motion signals in the chest, abdomen and limb regions are realized.

[0059] Further, step S1000 includes:

[0060] Step S1100: Construct an adaptive space beamforming operator based on space topology compensation for the acquired raw radar data stream. Perform complex weighted summation to generate a focused echo stream for a specific anatomical region. .

[0061] Specifically, this step aims to treat target echoes under multipath scattering conditions as the object of digital processing. Utilizing the spatial geometric distribution characteristics of the 76-81 GHz band frequency-modulated continuous wave radar and MIMO antenna array, the acquired raw radar data stream is processed. The multi-channel components are mapped as complex weighted vectors representing spatial directivity. By performing phase coherent superposition of the signals from each channel, the chaotic wide beam is reconstructed into a narrow-direction virtual probe targeting specific anatomical regions such as the myocardial motion region in the chest cavity, the respiratory fluctuation region in the abdomen, and the arterial pulsation region in the extremities. The focused echo stream after energy enhancement is then extracted. This eliminates interference components caused by bedding obstruction and environmental reflection at the data source.

[0062] In the specific implementation process, at the front-end data acquisition layer of the millimeter-wave radar terminal, a MIMO antenna array consisting of several sets of transmitting and receiving antennas is constructed using time-division multiplexing technology. A virtual spatial array of equivalent sampling channels. This millimeter-wave radar terminal captures the... Each equivalent sampling channel at the sampling time raw radar data stream .in, This represents the total number of equivalent sampling channels, which is determined by the number of virtual array elements generated by time-division multiplexing technology. Increasing the total number of equivalent sampling channels... The value of can be used to achieve more precise spatial division of different anatomical regions of the human body.

[0063] Because different anatomical regions of the human body, such as the chest wall displacement area corresponding to heart failure early warning, the abdominal area corresponding to respiratory and fluid monitoring, and the peripheral arterial area corresponding to cerebral infarction early warning, have azimuth angles relative to the spatial coordinates of the antenna array. Pitch angle and target distance Due to the difference in reflection path length, the electromagnetic waves radiated by the transmitting antenna are reflected by the surface of the human target and then synchronously captured by the receiving elements in the receiving antenna array. The phase shift of the electromagnetic waves is caused by the spatial difference in their arrival at each receiving element.

[0064] Further, please refer to Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the principle of this invention, from the MIMO antenna array to a non-anatomical region of the human body. (See diagram below.) Figure 2 As shown, a spatial coordinate system for the antenna array is established at the MIMO antenna array of the millimeter-wave radar terminal, defining the azimuth, elevation, and distance of the human target relative to the origin. Different anatomical regions of the human target, such as the thoracic myocardial motion region corresponding to heart failure early warning, the abdominal respiratory fluctuation region corresponding to respiratory and body fluid monitoring, and the peripheral arterial pulsation region corresponding to cerebral infarction early warning, have different specific anatomical positions relative to the origin, corresponding to different azimuth, elevation, and distances. Figure 2 The example demonstrates the azimuth, pitch, and target distance of the thoracic myocardial motion zone.

[0065] To decouple weak physiological signals within three specific monitoring spaces—the thoracic myocardial motion zone, the abdominal respiratory fluctuation zone, and the peripheral arterial pulsation zone—this step constructs an adaptive spatial beamforming operator based on spatial topology compensation within the digital processing unit. This adaptive spatial beamforming operator is essentially an adaptive beamforming algorithm based on spatial geometric constraints, which, through… The discrete signals from each equivalent sampling channel are summed using complex weights to generate a focused echo stream that reflects the vibration characteristics of a specific anatomical location. The adaptive characteristic is reflected in its ability to dynamically adjust spatial filtering performance based on real-time environmental feedback, while spatial topology compensation corrects phase shift by calculating the geometric travel difference between the antenna array and various organs of the human body.

[0066] The focused echo stream Focused echo intensity sequences with different spatial orientations at each sampling time The components are combined sequentially, and the specific execution logic is as follows: Each equivalent sampling channel is sampled at the sampling time... Raw radar data components and complex weighted vectors with spatial orientation characteristics Multiply, and over all The product results of the equivalent sampling channels are summed synchronously to obtain the sampling time. Below the azimuth angle and pitch angle Upward-pointing focused echo intensity sequence Wherein, the complex weighted vector The amplitude gain term is used to adjust the signal amplitude. and used to correct phase shift The product of phase compensation terms constitutes the equation.

[0067] The amplitude gain term The value is determined in the digital processing unit based on a preset Chebyshev window. The selection principle is to dynamically adjust the signal-to-clutter ratio (SCR) based on the real-time calculation of the echo intensity of the target area and the clutter intensity of the surrounding non-target area by the millimeter-wave radar terminal, thereby reducing the masking effect of environmental reflections on weak physiological signals. The phase compensation term... Phase correction factor It is based on the first MIMO antenna array The geometrical spatial difference between the spatial geometric coordinates of each physical array element and the coordinates of the center of the anatomical region of the target individual is determined; Represents the imaginary unit; This represents the index number of the equivalent sampling channel, with a value ranging from 1 to the total number of equivalent sampling channels. .

[0068] Step S1200, based on the collected target distance and azimuth Construct a logical decision matrix and use a spatiotemporal quadratic mapping function to analyze the focused echo flow. The signal components falling within the corresponding spatial thresholds of the chest, abdomen, and limbs are extracted and reconstructed to obtain the signal from the thoracic cavity channel. Abdominal passage and peripheral channels Composition of regional synchronization signal set .

[0069] Specifically, this step aims to focus the echo stream from step S1100. As a data mapping object, it utilizes digitally defined distance resolution units. and angle resolution window The constructed two-dimensional logical mask will focus the echo flow. The study maps the chest wall micromotion characteristics that characterize cardiac pumping dynamics, the abdominal undulation characteristics that characterize respiratory movements and fluid distribution, and the limb vascular pulsation characteristics that characterize peripheral circulatory pressure. This mapping achieves digital reconstruction and orthogonal decoupling of physiological characteristic information at the data source, generating data from the thoracic cavity channel. Abdominal passage and peripheral channels A regional synchronization signal set composed of three channels .

[0070] In specific implementation, the millimeter-wave radar terminal is based on a spatially directional focused echo stream. By constructing a spatiotemporal quadratic mapping function based on spatial geometric coordinate constraints, a multidimensional virtual detection channel is established in the memory buffer of the digital processing unit. The specific construction logic of the spatiotemporal quadratic mapping function is as follows: First, the target centroid coordinates are retrieved and locked. Establish a system with the target's centroid coordinates as the center and a range resolution unit. Window with radial width and angular resolution This is the logical decision matrix for the angular width. The logical decision matrix, by defining the data extraction boundary in the digital coordinate system, is used to determine the subsequent focused echo flow. The physical attribution is then determined. Subsequently, the spatiotemporal quadratic mapping function, by executing a matrix slicing operator based on spatial coordinate indices, strips and maps echo sequences falling within the coverage area of ​​the logical decision matrix to independent virtual monitoring channel storage areas. The distance resolution unit... The value is based on the target distance in step S1110. The established distance gate width is used to extract signal slices at a specific physiological monitoring depth along the radial distance dimension; the angular resolution window The value is based on the azimuth angle in step S1110. The established spatial resolution span is used to ensure the regional synchronization signal set. Each channel component contains only reflection information within the boundaries of a preset functional area, achieving orthogonal isolation in the airspace.

[0071] The spatiotemporal quadratic mapping function will focus the echo flow. Transformed into a regional synchronization signal set Its specific execution logic is as follows: Focus the input echo stream Perform range resolution unit and angle resolution window The logical mask operation; the logical mask operation constructs a Boolean decision operator in the memory buffer corresponding to the spatial thresholds of specific anatomical regions of the chest, abdomen, and limbs, for the focused echo stream. The discrete sampling point signal components are retrieved one by one: when the spatial location information of the sampling point meets both the distance threshold and the angle threshold, the Boolean decision operator assigns it a logically valid identifier, allowing the signal component to enter the corresponding virtual monitoring channel storage area through the mapping logic executed by the matrix slicing operator; otherwise, it is assigned a logically invalid identifier and masked, thereby suppressing clutter in non-target areas and motion interference from other parts of the human body; through the logical mask operation, the discrete signal components falling within the three specific spatial thresholds of the chest, abdomen, and limbs are extracted and recombined, ultimately generating mutually orthogonal chest channels in the memory space. Abdominal passage and peripheral channels The regional synchronization signal set The distance threshold is defined as the target distance. Extending radially to both sides by half the distance resolution unit from the center. The resulting radial distance interval; the angle threshold is in azimuth angle. Centered on the center, the window expands by half an angular resolution to both sides. The angular span range formed; the thoracic passage The target chest movement center coordinate range is used to extract displacement data caused by cardiac mechanical pulsation; the abdominal channel This refers to the target abdominal functional dynamic coordinate range, used to capture the undulating characteristics of respiratory drive and the undulating phase of the abdominal major blood vessels; the peripheral limb channels It is the coordinate range of the limb extremities, used to capture weak arterial pulsation signals caused by peripheral circulatory pressure waves.

[0072] Step S2000: Set the area synchronization signal. Mapped to a set of modal components And input it into a deep residual network to generate a heartbeat interval that is included in each beat. Atrial fibrillation probability score and pulse wave conduction time Physiological feature vector Based on regional synchronization signal set Fluid retention index was obtained from the abdominal and thoracic phase data sequences. .

[0073] Specifically, this step aims to integrate the regional synchronization signal set from step S1200, which includes dynamic displacement characteristics of the chest, abdomen, and limb regions. As the data processing object, the potential physiological micro-motion features in each channel, which are masked by sensory dynamic interference, are mapped into a set of modal components characterizing cardiac mechanical pulsation, respiratory rhythm, and end-circulatory fluctuations. The orthogonal subband components in the model are combined with the modal component set through a deep residual network. Modal component features are transformed into physiological feature vectors representing cardiac rhythm variability and circulatory status. Simultaneously, the phase shift characteristics in the abdominal region caused by changes in dielectric constant are mapped to a fluid retention index. This enables the extraction and expression of pathological information.

[0074] Further, step S2000 includes:

[0075] Step S2100: Use the variational mode decomposition operator to set the regional synchronization signal. The mapping is the sum of a finite number of narrowband mode functions. The objective function is to minimize the sum of the estimated bandwidths of each narrowband mode function. Global minimization is then performed to obtain the modal component set. .

[0076] Specifically, this step aims to integrate the regional synchronization signal set from step S1200. As a digital processing object, the frequency domain distribution differences of vibration energy in different physiological functional zones on the complex analytic plane are utilized to construct a variational mode decomposition operator to analyze the thoracic cavity passage. Abdominal passage and peripheral channels Internally mixed non-stationary sequences are mapped to a set of modal components with specific center frequencies. It achieves orthogonal separation of cardiovascular pulsation characteristics from environmental clutter and random bodily movements in the digital frequency domain, and preserves the phase evolution characteristics between cross-regional signals while suppressing random interference outside the preset physiological frequency band.

[0077] In practice, due to the complex and multi-interference monitoring scenarios, the chest wall micro-vibrations, abdominal undulations, and peripheral pulse signals of the monitored individuals exhibit non-stationary and non-linear dynamic characteristics, and are highly susceptible to digital masking by high-frequency noise such as environmental mechanical simple harmonic vibrations or limb micro-swaying. To achieve accurate extraction of physiological motion information, this step establishes a variational mode decomposition operator and uses a preset regional synchronization signal set. Each channel signal in the algorithm is the sum of a finite number of narrowband mode functions with specific center frequencies. The objective function is to minimize the sum of the square norms of the estimated bandwidths of each mode component, thereby maximizing the suppression of cluttered background components in the digital domain. Subsequently, the alternating direction multiplier method is used as a numerical solution strategy. Within the memory space of the digital processing unit, the optimal solution set that minimizes the objective function is found through alternating iterations. This decouples a set of narrowband mode functions that can characterize the original physiological rhythms, i.e., the modal component set. .

[0078] The modal component set The specific generation logic is as follows: A constraint objective function is established using the variational mode decomposition operator; the analytic signal is obtained by performing a Hilbert transform on the narrowband mode functions of each channel signal; and the spectrum of the analytic signal is shifted to the baseband using center frequency shift processing, thereby obtaining an analytic sequence reflecting the evolution of the signal envelope; subsequently, the gradient of this analytic sequence is calculated. The norm square is used to characterize the estimated bandwidth of each modal component; finally, the alternating direction multiplier method is used to find the minimum of the constrained objective function, so that the sum of the estimated bandwidths of each modal component is minimized, and finally the signal set is synchronized from the noisy region. Extract mutually orthogonal modal components with stable center frequencies from the data. This allows for the separation of physiological motion information from environmental background noise at the digital source. The center frequency, defined as the digitized center frequency value of each modal component, is automatically updated during the iteration process using an alternating direction multiplier method based on the signal energy distribution. This is used to lock the energy concentration band of the physiological motion signal, achieving adaptive alignment of the signal in the frequency dimension.

[0079] Step S2200: Use a deep residual network to set the modal components. The mid-thoracic modal components are mapped to beat-by-beat heart rate intervals. The modal components of the chest cavity, abdomen, and peripheral extremities are mapped to atrial fibrillation probability scores. and the thoracic passage and peripheral channels The spatiotemporal cross-correlation peak time delay is mapped to the pulse wave propagation time. Generate physiological feature vectors .

[0080] Specifically, this step aims to extract the modal component set from step S2100. As a digital reconstruction object, the skip connection operator of deep residual networks is used to nonlinearly enhance the weak physiological vibration characteristics that are masked by disturbances, thereby revealing the underlying modal component set. The thoracic modal components are mapped to beat-by-beat heart rate intervals that characterize the features of heart rhythm variability. modal component set The thoracic, abdominal, and peripheral modal components are mapped to an atrial fibrillation probability score characterizing embolism risk. and the thoracic passage and peripheral channels Spatiotemporal correlation is mapped to pulse wave propagation time, which characterizes vascular compliance. The simultaneous extraction and vectorization of multiple pathological parameters are achieved in the feature space to obtain a result containing the beat-by-beat heartbeat interval. Atrial fibrillation probability score and pulse wave conduction time Physiological feature vector .

[0081] In the specific implementation process, this step establishes a deep residual network based on a multi-task learning architecture to process the modal component set output from the previous steps. In complex and multi-interference monitoring scenarios, sub-millimeter-level skin micro-displacement data caused by cardiac contraction are prone to vanishing feature gradients during deep residual learning, resulting in weakened digital representation capabilities.

[0082] This step introduces a skip connection operator between the convolutional layers of the residual block, allowing the modal components to be reconstructed entering the residual block to be linearly superimposed directly with the mapped feature components extracted by the convolutional layers, bypassing the nonlinear transformation layers. This ensures data integrity during the multi-level nonlinear mapping process. A deep residual network is then used to process the modal component set. Iterative transformation is performed on the modal components of each channel to generate physiological feature vectors. Among them, the physiological feature vector It includes the heart rate interval per beat. Atrial fibrillation probability score and pulse wave conduction time A multidimensional numerical set is used to quantify the real-time cardiovascular risk status of a target individual; the beat-by-beat heart rate interval It is a physiological feature vector The first component, whose value originates from the modal component set. The mesothoracic modal component is used to characterize the time-series variability caused by myocardial pulsation; the atrial fibrillation probability score... It is a physiological feature vector The second component, whose value originates from the modal component set. The modal components of the mid-thoracic cavity, abdomen, and peripheral extremities, taking values ​​ranging from 0 to 1, are used to assess the probability of cardioembolic risk; the pulse wave conduction time... It is a physiological feature vector The third component, whose value originates from the modal component set. Midthoracic passage and peripheral channels The peak time delay of the spatiotemporal cross-correlation of the signal is used to quantify the millisecond-level time delay between the point of cardiac mechanical contraction sensed by electromagnetic waves and the point of arrival of peripheral pulses, serving as a characteristic value for assessing vascular compliance.

[0083] Further, please refer to Figure 3 As shown, Figure 3 This is a schematic diagram illustrating the principle of physiological feature vector extraction based on monitoring different anatomical regions of the human body according to the present invention. For example... Figure 3 As shown, a deep residual network is used to process the input modal component set. Perform multipath mapping on each modal component set. First path: map the modal component sets... The thoracic modal components are mapped to beat-by-beat heart rate intervals that characterize the features of heart rhythm variability. Second path: The modal component set The thoracic, abdominal, and peripheral modal components are collectively mapped to an atrial fibrillation probability score that characterizes the risk of embolism. Third path: The modal component set The spatiotemporal correlation between the thoracic cavity and peripheral limb channels is mapped to the pulse wave conduction time, which characterizes vascular compliance. Finally, the above-mentioned beat-by-beat heart rate intervals Atrial fibrillation probability score and pulse wave conduction time Vectorization and recombination are performed to generate physiological feature vectors for quantifying the real-time cardiovascular risk status of target individuals. .

[0084] Step S2300: During the monitoring period, retrieve the abdominal channel data. Abdominal phase data sequence and thoracic passage The chest phase data sequence is used, and the absolute value of the difference between the abdominal and chest phase data sequences is integrated over time. Combined with a preset attenuation compensation factor and static reflection reference intensity, a fluid retention index is generated. .

[0085] Specifically, this step aims to utilize the abdominal channel generated in step S1200. With the thoracic passage As a digital processing object, utilizing the physical properties of the interaction between electromagnetic waves and biological tissue components, the evolution characteristics of the complex dielectric constant of tissues caused by pathological fluid accumulation within the monitoring area are mapped into a quantitative sequence characterizing the degree of phase deviation. In the digital logical space, the analysis and separation of information from the monitored anatomical region, tissue electromagnetic scattering properties, and tissue dielectric properties are achieved, generating a fluid retention index characterizing the evolution of the circulatory system load. This provides real-time feature data input for subsequent steps to identify the cross-feedback between heart failure and cerebral infarction risk.

[0086] In the specific implementation process, this step retrieves the abdominal channel. With the thoracic passage The discrete phase data sequences within the monitoring period include abdominal phase data sequences and thoracic phase data sequences. During the pathological evolution of heart failure, the accumulation of body fluids in the abdominal and lower limb tissues causes a shift in the complex permittivity of the monitored anatomical region relative to electromagnetic waves in the 76-81 GHz band, which in turn causes an instantaneous phase drift of the reflected wave on the time axis.

[0087] To extract pathological features, this step executes phase difference offset logic in the memory buffer, using the thoracic cavity channel. As a physiological benchmark, it cancels common-mode interference caused by non-targeted physical disturbances or respiratory baseline drift in the detection scene, calls the dynamic operator of body fluid reflection coefficient, and performs time-domain accumulation calculation on the relative phase deviation of the abdomen and thoracic cavity, thereby establishing a deterministic mapping relationship from physical field phase shift to digital pathological indicators.

[0088] The specific execution logic of the dynamic operator for the body fluid reflection coefficient is as follows: body fluid retention index This is equivalent to, within the monitoring period, the sampling time The absolute value of the difference between the instantaneous abdominal phase of the abdominal phase data sequence and the instantaneous thoracic phase of the thoracic phase data sequence is integrated over time and divided by the product of the attenuation compensation factor and the static reflection reference intensity. The monitoring period is a preset continuous monitoring time window width, covering the nighttime static sampling period of the target individual. The attenuation compensation factor is a fixed correction coefficient preset based on the dielectric loss characteristics of the monitoring environment, used to eliminate non-pathological interference of physical barriers on the electromagnetic field phase. The static reflection reference intensity is a preset human static electromagnetic reflection reference intensity value, determined during the initial calibration phase by measuring the reflectivity of non-edema tissues of the target individual in a physiologically stable state, ensuring the comparability of indicators between different individuals.

[0089] Step S3000, based on physiological feature vector and fluid retention index Generate a linkage feature matrix by executing the heart failure-stroke correlation mapping logic. Based on the linkage feature matrix Execute dangerous pattern recognition logic to generate interaction enhancement coefficients .

[0090] Specifically, this step aims to transform the physiological feature vector from step S2200... and the fluid retention index from step S2300 As a data processing object, the pathological coupling characteristics of cardiac power output and peripheral circulatory feedback are utilized to map the latent pathological synergistic trends at the microscopic level, which are difficult to characterize with single parameters, into a linkage feature matrix. Furthermore, it achieves nonlinear interactive enhancement of the risk weights of heart failure and cerebral infarction during specific risk enhancement pattern recognition processes, and outputs an interactive enhancement coefficient representing the synergistic risk level. .

[0091] Further, step S3000 includes:

[0092] Step S3100, based on physiological feature vector and fluid retention index The heart failure-stroke correlation mapping logic is executed to generate a linkage feature matrix containing covariance components, vascular resistance evolution components, arrhythmia components, and nonlinear enhancement components. .

[0093] Specifically, this step aims to transform the physiological feature vector from step S2200... and the fluid retention index characterizing volume overload from step S2300 As a multidimensional digital mapping object, it utilizes the spatiotemporal correlation characteristics of heart failure and cerebral infarction in the pathological evolution process to calculate the probability score of atrial fibrillation in cardiac pumping function. heart rate interval per beat and the fluid retention index, which reflects the evolutionary characteristics of fluid accumulation in peripheral tissues. Mapped to a linkage feature matrix The nonlinear coupling signal in the data enables the correlation modeling of dynamic indicators and capacity load indicators.

[0094] In the specific implementation process, since atrial fibrillation-induced blood turbulence is the physical trigger for thrombus formation in the evolution of cardiovascular events, and the increased volume load caused by heart failure worsens intracardiac pressure and accelerates the detachment of potential thrombi, this step constructs a heart failure-cerebral infarction correlation mapping logic, and no longer uses physiological feature vectors. and fluid retention index Treated as an isolated variable. This heart failure-stroke association mapping logic identifies the fluid retention index. The increased volume overload, as characterized by this phenomenon, leads to a deterioration in intracardiac pressure, which in turn accelerates the physiological characteristic vector. atrial fibrillation probability score The correlation patterns of potential thrombus detachment were characterized, and the linkage of cross-regional features was constructed.

[0095] The specific mapping logic of the heart failure-cerebral infarction correlation mapping logic is as follows: The physiological feature vector... atrial fibrillation probability score and fluid retention index The covariance operator, i.e., the covariance components, within the monitoring period serve as the linkage feature matrix. The first element of the main diagonal is used to quantify the risk enhancement effect; the physiological feature vector is used... Mid-pulse wave conduction time For sampling time The product of the first derivative and the preset first weighting coefficient is used as the vascular resistance evolution component characterizing peripheral vascular resistance, i.e., as the linkage feature matrix. The first row and second column element; the physiological feature vector Mid-beat heart rate variability The product of the reciprocal of the product and the preset second weighting coefficient is used as the arrhythmia component characterizing the degree of arrhythmia, i.e., as the linkage feature matrix. The second row and first column elements; and the residual correlation operator is introduced to extract physiological feature vectors. and fluid retention index The implicit collaborative features between them are used as the product of the implicit collaborative features and the preset third weighting coefficient as the linkage feature matrix. The nonlinear enhancement component is used as the linkage feature matrix. The main diagonal tail element. Wherein, the linkage feature matrix... It characterizes the degree of coupling between pathological features in different anatomical regions of a target individual. Digital mapping set; the covariance operator is used to extract atrial fibrillation probability score. and fluid retention index The spatiotemporal coupling correlation characteristics between these factors are used to determine the degree of synergistic deterioration between the risk of cardioembolic stroke and the circulatory system volume load within the sampling period. The values ​​of the preset first weighting coefficient, preset second weighting coefficient, and preset third weighting coefficient are fixed values ​​preset based on the target individual's basic medical history records, and are used to dynamically adjust the correlation features of different pathological characteristics within the linkage feature matrix. The weighting percentage in the data; the variation in heart rate intervals per beat. The value is derived from the sampling time. heartbeat interval between two adjacent sampling periods The difference is used to quantify the unstable state of the heart rhythm at a specific sampling time scale.

[0096] The residual correlation operator is a mapping function used to extract nonlinear collaborative features within the digital pathological feature interaction mapping domain, and is used to identify latent pathological associations. Its specific execution logic is as follows: First, the least squares linear regression method is used as a detrending algorithm to strip away the physiological feature vectors. and fluid retention index The linear trend term in the data yields two residual sequences representing weak perturbations: a physiological characteristic residual sequence and a fluid retention residual sequence. Subsequently, the Pearson correlation coefficients of the physiological characteristic residual sequence and the fluid retention residual sequence are calculated within the monitoring period to generate a sliding cross-correlation coefficient sequence reflecting the instantaneous association strength. The sliding cross-correlation coefficient sequence is then subjected to time averaging to extract its latent pathological associations within the monitoring period, i.e., latent synergistic features.

[0097] Step S3200, based on the linkage feature matrix The hazard pattern recognition logic is executed, and a risk probability mapping value is generated by combining it with a normalized exponential function. The risk probability mapping value is then multiplied by a nonlinear enhancement exponent raised to the power of a preset risk coupling constant to obtain the interaction enhancement coefficient. .

[0098] Specifically, this step aims to transform the linked feature matrix from step S3100. As a digital decision-making object, it utilizes the pathological co-evolutionary characteristics between cardiac power output and circulatory volume load to link the feature matrix. The underlying, single-parameter-difficult-to-characterize pathological risk evolution patterns are mapped to the interaction enhancement coefficient. It achieves quantitative enhancement of latent risks in the feature processing dimension, captures potential crises caused by the synergistic deterioration of multidimensional pathological indicators, and provides digital data support with pathological logic correction capabilities for the subsequent generation of early warning decision instructions.

[0099] In the specific implementation process, this step retrieves the linkage feature matrix generated by the previous step. Based on the aforementioned linkage feature matrix Based on the characteristic distribution patterns of each component, execute the dangerous pattern recognition logic and analyze the linked feature matrix. Nonlinear feature combination analysis is performed on the correlated components within the matrix. This hazard pattern recognition logic is used to capture risk combination analysis with pathological synergistic deterioration characteristics, for example, when a linked feature matrix is ​​detected. The covariance component characterizing increased atrial fibrillation burden and the fluid retention index characterizing [the following components are used in the text]. When the nonlinear enhancement components showing an increasing trend together constitute the preset evolution characteristics, it is determined that the probability of thrombus detachment and entry into the systemic circulation due to changes in cardiac pressure is nonlinearly increasing, rather than being a linear sum of various risk indicators.

[0100] The specific execution logic of the dangerous pattern recognition logic is as follows: First, the linkage feature matrix is... Each linked feature component, namely the covariance component, the vascular resistance evolution component, the arrhythmia component, and the nonlinear enhancement component, is mapped and multiplied with its corresponding hazard mode weight, and then summed to obtain a feature aggregation value. Subsequently, the feature aggregation value is projected onto a preset distribution interval using a normalized exponential function to obtain a preliminary risk probability mapping value. Finally, the risk probability mapping value is multiplied with a nonlinear enhancement exponent raised to the power of the risk coupling constant to obtain the interaction enhancement coefficient. This allows for exponential reinforcement of latent risks at the digital processing terminal. The hazard pattern weights are vector sets constructed from prior clinical pathology data, containing multiple sets of high-risk pathological feature templates. The specific pre-setting method involves: based on the early warning monitoring dimensions of the target individual, linking the feature matrix... The four linked feature components are assigned proportional constants that are positively correlated with their pathological contributions to form a weight vector. For example, for the high-risk pathological feature template of "atrial fibrillation with volume surge," the weight values ​​corresponding to the covariance component and the nonlinear enhancement component are increased, such as setting the weight coefficient to 0.6-0.8, while the weight values ​​of the vascular resistance evolution component and the arrhythmia component are decreased, such as setting the weight coefficient to 0.1-0.2. The normalization exponential function is the Softmax function, used to project the weighted aggregated feature values ​​onto... The probability distribution space ensures the interaction enhancement coefficient. Within a preset effective range, the risk coupling constant is a preset fixed value used to control the rate of risk enhancement. Its value is determined offline based on the target individual's clinical history risk level. Specifically, based on whether the target individual has a history of atrial fibrillation, hypertension, or previous heart failure, the corresponding value is retrieved from a preset risk constant mapping table. The higher the clinical history risk level, the larger the preset risk coupling constant value, thereby increasing the interaction enhancement coefficient. Sensitivity to multi-parameter synchronous fluctuations; when the linkage feature matrix When multiple linked feature components exhibit a synchronous deterioration trend, this constant achieves a nonlinear mapping of risk levels through an exponential effect, capturing high-risk intervals where individual indicators are within a preset safe range but the synergistic state is abnormal. The preset risk constant mapping table is a multidimensional numerical index matrix pre-stored in the processing terminal. By combining and quantifying clinical risks such as atrial fibrillation, hypertension, and a history of heart failure, it delineates corresponding risk level boundaries and matches them with appropriate constant values. For example, for individuals without relevant medical history, the risk constant mapping table outputs a lower risk coupling constant; while for high-risk individuals with both a history of atrial fibrillation and heart failure, the risk constant mapping table assigns a higher risk coupling constant, such as a scalar value above 2.0. The nonlinear enhancement index is an exponential mapping operator with a natural constant as its base, used to introduce an exponential growth characteristic of risk intensity into the hazard pattern recognition logic.

[0101] Step S4000, based on physiological feature vector and linkage feature matrix Execute a dual-path risk assessment logic to generate independent risk score pairs, combined with an interaction enhancement coefficient. Implement digital integration logic to obtain a comprehensive risk assessment value. And using segmented threshold discrimination logic to determine the comprehensive risk assessment value Mapped to tiered early warning decision instructions .

[0102] Specifically, this step aims to transform the physiological feature vector from step S2200... The linkage feature matrix from step S3100 and the interaction enhancement coefficient from step S3200 As a comprehensive treatment object, the nonlinear superposition effect of heart failure and cerebral infarction in the pathological evolution is utilized to analyze the physiological feature vectors in a multidimensional heterogeneous state. and linkage feature matrix Mapped to independent risk score pairs And combined with the interaction enhancement coefficient The independent risk score Probabilistic fusion is performed to generate a comprehensive risk assessment value that characterizes the overall cardiovascular risk status. To execute the corresponding energy level-based hierarchical early warning decision instructions Provides real-time quantitative data support.

[0103] Further, step S4000 includes:

[0104] Step S4100, based on the linkage feature matrix and physiological feature vectors Execute the dual-path risk assessment logic to generate a heart failure risk index. and stroke risk index Independent risk score pairs.

[0105] Specifically, this step aims to transform the physiological feature vector from step S2200... and the linkage feature matrix from step S3100 characterizing the cross-regional pathological coupling strength As a parallel computing object, leveraging the synchronous evolution of heart failure and cerebral infarction risks in the digital data domain, it executes dual-path risk assessment logic to process data including beat-by-beat heart rate intervals. Atrial fibrillation probability score and pulse wave conduction time Physiological feature vector Mapped to a characterizing index of heart failure risk and stroke risk index The independent risk score pairs enable quantitative grading of single lesion risk and collaborative quantitative calibration of cross-regional lesion risk in the dimension of digital processing.

[0106] In the specific implementation process, this step constructs a dual-path risk assessment logic based on a multi-task parallel processing architecture. This dual-path risk assessment logic retrieves the heartbeat interval output from the previous step. Atrial fibrillation probability score and pulse wave conduction time Physiological feature vector and the linkage feature matrix characterizing the strength of pathological correlation. In terms of processing logic, this step utilizes the aforementioned linkage feature matrix. The covariance component and nonlinear enhancement component in the vector are used as dynamic calibration factors for the nonlinear projection, applied to the physiological feature vector. Real-time weight adjustments are performed on each feature component. This parallel computing path ensures that while identifying cardiogenic pathological changes, peripheral circulatory response data regulated by volume load can be acquired simultaneously, thereby enabling parallel discrimination of heart failure and cerebral infarction risk within the digital domain.

[0107] The dual-path risk assessment logic performs iterative transformations using a nonlinear feature mapping operator to achieve parallel quantification of risk indicators, mapping them respectively to heart failure risk indices. and stroke risk index The two risk indices are encapsulated into independent risk scoring pairs. This dual-path risk assessment logic includes a heart failure risk assessment pathway and a stroke risk assessment pathway.

[0108] The heart failure risk assessment pathway integrates the physiological feature vectors. The beat-by-beat heartbeat interval as a time series component And the fluid retention index, which is a characteristic of body fluid distribution. Calculate and generate a heart failure risk index The specific digital calculation formula for this path branch is expressed as follows: the heart failure risk index This is equivalent to the heart rate interval per beat. Fluid retention index and heart rate variability As a first set of feature terms, after performing a first nonlinear feature mapping operator transformation on the first set of feature terms, it is then aggregated and summed with a preset first feature weight coefficient. Wherein, the heart rate variability... The value is derived from the heart rate interval per beat. Statistical analysis of standard deviation within a preset monitoring window is used to quantify the regulatory function of the autonomic nervous system on the heart; the first nonlinear feature mapping operator and the second nonlinear feature mapping operator in the subsequent stroke risk assessment path are used to realize the digital conversion of physiological features from the original signal domain to the pathological risk domain; the values ​​of the first feature weight coefficient and the second feature weight coefficient in the subsequent stroke risk assessment path are affected by the linkage feature matrix. The real-time correction of the dynamic calibration factor is used to dynamically adjust the contribution intensity of each feature component to risk assessment.

[0109] The specific execution logic of performing the first nonlinear feature mapping operator transformation is as follows: The components of the first feature term set are mapped to using the deviation normalization algorithm. The standard scale of the interval is used to obtain a standardized feature sequence, eliminating dimensional differences between different physiological parameters. Then, a nonlinear activation function, such as the sigmoid function, is used to perform numerical projection on the standardized feature sequence, transforming the linear fluctuations of the physiological signal into a sensitivity response component characterizing pathological risk. This sensitivity response component is then projected onto the pathological risk domain using a preset mapping matrix, generating a first nonlinear vector characterizing the features of cardiogenic lesions. The preset mapping matrix is ​​a weight transformation operator pre-stored in the processing terminal, whose parameter values ​​are obtained through offline regression analysis based on prior clinical pathological data, used to quantify the contribution weight of each feature item to pathological risk. The pathological risk domain is a digital quantification space used to measure the degree to which the physiological state deviates from the normal threshold. Through the projection operation of the preset mapping matrix, the sensitivity response component is transformed into a risk feature vector characterizing a specific lesion trend, i.e., the first nonlinear vector.

[0110] The stroke risk assessment pathway simultaneously utilizes the aforementioned physiological feature vectors. Atrial fibrillation probability score as a component of atrial fibrillation burden And pulse wave propagation time as a component of vascular compliance Calculate and generate a stroke risk index The specific digital calculation formula for this path branch is expressed as follows: the stroke risk index Equivalent to a score including the probability of atrial fibrillation Pulse wave conduction time and blood pressure variability As a second set of feature terms, after performing a second nonlinear feature mapping operator transformation on the second set of feature terms, it is then aggregated and summed with preset second feature weight coefficients. Among these, the blood pressure variability... The value is derived from the pulse wave conduction time. The fluctuation dispersion calculation within a preset time window is used to quantify the physical driving effect of peripheral circulatory pressure fluctuations on the dissection of unstable thrombi. The execution logic of performing the second nonlinear feature mapping operator transformation is consistent with the logic of performing the first nonlinear feature mapping operator transformation.

[0111] Step S4200: For independent risk score pairs and interaction enhancement coefficients... Implementing digital integration logic, the heart failure risk index The product of the preset first fusion compensation coefficient and the stroke risk index The product of the preset second fusion compensation coefficient and the interaction enhancement coefficient. The product of the product and the preset third fusion compensation coefficient is summed to generate a comprehensive risk assessment value. .

[0112] Specifically, this step aims to combine the independent risk score pairs generated from the dual-path risk assessment logic in step S4100 with the interaction enhancement coefficient representing the cardiocerebrovascular risk linkage enhancement effect from step S3200. As a multi-dimensional data fusion object, using digital fusion logic, the heart failure risk index is... and stroke risk index We perform aggregated reconstruction at the pathological mechanism level and introduce an interaction enhancement coefficient. This achieves nonlinear compensation for implicit collaborative risks at the edge of independent alarm thresholds, mapping the discrete risks of a single disease to a globally consistent comprehensive risk assessment value within the digital feature space. This is to facilitate the subsequent execution of the tiered early warning strategy and the output of tiered early warning decision instructions. Provide numerical evidence supported by pathological logic.

[0113] In its implementation, this step no longer employs simple arithmetic averaging or linear summation. Instead, it establishes a digital fusion logic targeting the risk of heart-brain synergy. This digital fusion logic, under the same sampling benchmark, will characterize the risk index of heart failure with reduced pumping power. Stroke risk index, which characterizes the risk of thrombus detachment. and the interaction enhancement coefficient characterizing the intensity of pathological co-evolution. The process involves performing temporal feature spatial mapping and nonlinear alignment. By identifying the pathological co-evolution cycles of different disease risk pathways within the monitoring period, this process captures the risk of acute decompensation of circulatory function caused by atrial fibrillation-induced embolus detachment and abnormal synchronization of circulatory volume load.

[0114] The digital fusion logic is used to quantify the overall risk level and calculate a comprehensive risk assessment value. The specific execution logic is as follows: the comprehensive risk assessment value Equal to the risk index of heart failure The product of the first fusion compensation coefficient and the stroke risk index The product of the preset second fusion compensation coefficient and the interaction enhancement coefficient. The sum of the products of the comprehensive risk assessment value and the preset third fusion compensation coefficient. It is a standardized real value characterizing the global cardiovascular safety level of the monitored individual within the current sampling period, used as the sole digital basis for triggering the Level 3 early warning decision instruction; the preset first fusion compensation coefficient and the preset second fusion compensation coefficient are real numbers ranging from 0 to 1, and their values ​​are determined based on the target individual's basic physiological reflex benchmarks, such as a history of myocardial infarction or atrial fibrillation, to dynamically adjust the digital contribution ratio of the heart failure risk assessment path and the cerebral infarction risk assessment path in the comprehensive score; the preset third fusion compensation coefficient is usually set as a risk amplification constant greater than 1, and its value is derived from the linkage feature matrix characterizing the cross-regional pathological coupling strength in step S3100. Second derivative sensitivity analysis was performed to control the interaction enhancement coefficient. Total risk assessment value The nonlinear step gain intensity.

[0115] Step S4300, based on the comprehensive risk assessment value The segmented threshold discrimination logic is executed, and the comprehensive risk assessment value is determined by combining it with the preset early warning boundary threshold. The risk assessment range in which it is located is determined, and the output status bit is assigned the value of the first-level flag bit. Secondary flag position Or a level 3 flag Hierarchical early warning decision instructions .

[0116] Specifically, this step aims to integrate the comprehensive risk assessment value representing the intensity of heart-brain synergy risk from step S4200. As a digital assessment object, the comprehensive risk assessment value is determined using segmented threshold discrimination logic. Projection is a hierarchical early warning decision instruction containing discrete state bits. This enables the procedural transformation from multidimensional feature assessment to deterministic intervention actions, solving the problem of discontinuity of physiological signals caused by changes in the body position or physical obstruction of the monitored object in complex, non-intrusive monitoring environments, i.e., data discontinuity warning failure. It provides a digital decision-making basis with pathological logic for the early interception of the risk of heart failure decompensation and acute cerebral embolism.

[0117] In the specific implementation process, this step retrieves the comprehensive risk assessment value generated by the previous steps. By establishing a first fluctuation range within the instruction mapping domain Second abnormal interval and the third reinforcement zone The decision-making matrix, i.e., the risk assessment interval, serves as the mapping framework for executing the segmented threshold discrimination logic. This segmented threshold discrimination logic utilizes preset warning boundary thresholds to transform dynamically evolving pathological correlation trends into discrete machine-coded instructions, ensuring the comprehensive risk assessment value corresponding to each numerical state. The corresponding early warning response is unique. The decision-making matrix is ​​based on a comprehensive risk assessment value. and The quantitative logical relationship is used to programmatically determine the hierarchical early warning decision instructions. The status bit is assigned a value; the warning boundary threshold includes a first warning boundary threshold. Second warning boundary threshold And meets the first early warning boundary threshold. Less than the second warning boundary threshold The first warning boundary threshold Second warning boundary threshold The values ​​are all derived from the synchronous calibration of the monitored individual's cardiac power reserve benchmark and historical health event distribution, and are used to define the risk logic boundaries of different severity levels.

[0118] The specific execution logic of the segmented threshold discrimination logic is as follows: when determining the comprehensive risk assessment value... In the first fluctuation range At that time, i.e., the comprehensive risk assessment value Greater than or equal to 0 and less than the preset first warning boundary threshold At that time, a tiered early warning decision instruction will be issued. The status bit is assigned to the first-level flag bit. This triggers the generation of static health reports and the push of observation suggestions.

[0119] When determining the comprehensive risk assessment value Located in the second abnormal range At that time, i.e., the comprehensive risk assessment value Greater than or equal to the first warning boundary threshold And less than the second warning boundary threshold If the monitoring subject is deemed to have a latent risk arising from the co-evolution of heart failure and cerebral infarction risks, a tiered early warning decision will be issued. The status bit is assigned to the second-level flag bit. This triggered a notification for a follow-up outpatient visit.

[0120] When determining the comprehensive risk assessment value In the third enhancement zone At that time, i.e., the comprehensive risk assessment value Greater than or equal to the second warning boundary threshold At that time, it is identified as the cardiovascular system entering a decompensated critical mode, and a graded early warning decision instruction will be issued. The status bit is assigned to the level 3 flag bit. Immediately activate the emergency response mechanism, which includes push notifications to associated terminals and pre-activation of the emergency response terminal.

[0121] Step S5000, based on the comprehensive risk assessment value Generate baseline offset vector Combined with clinical feedback signals obtained from medical interactive terminals For the initial decision parameter set including fusion compensation coefficient and early warning boundary threshold Execute incremental adaptive calibration logic and output an adaptive decision parameter set. .

[0122] Specifically, this step aims to incorporate the comprehensive risk assessment value from step S4200. Using historical time series as a digital evolutionary driver, and leveraging the dynamic characteristics of the physiological characteristics of monitored individuals over time, the deviation of the current physiological risk state from a preset physiological homeostatic baseline is mapped as a baseline offset vector. And through incremental adaptive calibration logic, the adaptive decision parameter set, which includes fusion compensation coefficients and early warning boundary thresholds, is adjusted. By implementing closed-loop correction, dynamic alignment of individual-specific differences of monitored objects is achieved at the data processing layer, providing early warning benchmark support with spatiotemporal evolution characteristics for subsequent monitoring cycles.

[0123] Further, step S5000 includes:

[0124] Step S5100, according to the sampling time The timestamp index will retrieve the cumulative comprehensive risk assessment value from multiple sampling periods. The historical data stream is sorted into historical time series, and the static physiological baseline of the historical time series within a preset resting period is extracted. Based on instantaneous comprehensive risk assessment value and static physiological baseline Generate baseline offset vector .

[0125] Specifically, this step aims to store the comprehensive risk assessment value, which records the risk evolution trajectory of the monitored object in the digital memory, as described in step S4200. As a digital processing object, by constructing a historical time series reflecting the risk evolution trajectory of the monitored object, the current sampling time is... Instantaneous comprehensive risk assessment value Relative to static physiological baseline The numerical deviation is mapped to a baseline offset vector characterizing the intensity of the deviation from an individual's physiological state. .

[0126] In the specific implementation process, this step retrieves the comprehensive risk assessment value output from the previous steps and accumulated and stored in the digital memory over multiple sampling periods, recording the long-term risk evolution trajectory of the monitored object. Historical data stream, based on each sampling time The historical data stream is rearranged chronologically using timestamp indexes to form a historical time-series sequence reflecting the evolution of risk. Subsequently, this step executes spatiotemporal feature mapping logic, utilizing a sliding time window to identify and extract the mean value of the monitored object's sequence values ​​within a preset resting period. This preset resting period is the radar echo signal amplitude envelope data collected by the millimeter-wave radar terminal. By calculating the variance distribution of the radar echo signal amplitude envelope data in the time domain, the body motion energy density, representing the intensity of the monitored object's limb movement, is extracted. When the body motion energy density is lower than a preset steady-state threshold determined during the initial calibration phase, the monitored object is determined to be in deep sleep. The preset steady-state threshold is an energy limit value set based on the echo clutter level of the monitored object in a state of no body movement. By performing mean aggregation on the historical comprehensive risk assessment values ​​within the preset resting period, a static physiological baseline representing the distribution center of the individual's health characteristics is constructed. To ensure this static physiological baseline Non-pathological fluctuations caused by drastic changes in activity intensity or random electromagnetic interference in the environment were ruled out.

[0127] Subsequently, this step calculates the current sampling time. The generated instantaneous comprehensive risk assessment value With the static physiological baseline The difference in the spatial distribution of numerical values ​​between them generates a baseline offset vector. The baseline offset vector The specific generation logic is as follows: the baseline offset vector Equal to the sampling time within the total number of sampling points Instantaneous comprehensive risk assessment value and static physiological baseline The sum of squares of the differences is divided by the square root of the total number of sampling points, and then multiplied with the baseline adjustment operator. The total number of sampling points is the total number of sampling points within the sliding time window, and its value is preset as a positive integer based on the radar sampling frequency. The value of the baseline adjustment operator comes from a nonlinear mapping of body kinetic energy density, used to identify and eliminate baseline offset artifacts caused by non-pathological factors, i.e., abrupt changes in feature point clusters caused by non-pathological movements, ensuring the accuracy of offset quantification.

[0128] Step S5200, based on the comprehensive risk assessment value Clinical feedback signals obtained from medical interactive terminals Calculate the risk assessment deviation value and the baseline offset vector. Partial derivative operations are performed to obtain the feature offset gradient operator, which is then combined with a preset learning rate operator to form an initial decision parameter set containing the first, second, and third fusion compensation coefficients, as well as the warning boundary threshold. Perform incremental adaptive calibration and output an adaptive decision parameter set. .

[0129] Specifically, this step aims to transform the baseline offset vector from step S5100, which characterizes the evolution of the physiological characteristics of the monitored individual, into a vector that represents the evolution of the physiological characteristics of the monitored individual. and clinical feedback signals characterizing true pathological outcomes As a driver of parameter optimization, incremental adaptive calibration logic is used to map risk assessment deviation values ​​to an adaptive decision parameter set. This achieves closed-loop alignment between early warning criteria and individual physiological benchmarks in the dimension of digital processing.

[0130] In the specific implementation process, this step retrieves the baseline offset vector generated by the previous step through the communication link. Simultaneously acquire clinical feedback signals input from the medical interactive terminal. The clinical feedback signal The baseline offset vector serves as the true label value for incremental learning, used to quantify the residual distribution between the predicted value and the actual physiological state of the target individual. This step constructs an incremental adaptive calibration logic based on the stochastic gradient descent algorithm. This incremental adaptive calibration logic uses the baseline offset vector... The clinical feedback signal serves as the gradient direction for parameter search. As an adaptive decision parameter set The iterative convergence constraints construct a self-evolving, digital calibration and early warning decision logic for specific monitoring individuals.

[0131] The incremental adaptive calibration logic dynamically fine-tunes the weights of the calibration early warning decision logic by performing a loss function minimization operation, thereby adjusting the comprehensive risk assessment value generated in real time in step S4200. Clinical feedback signals acquired synchronously with medical interactive terminals Perform a correlation subtraction operation to obtain the comprehensive risk assessment value used as the prediction value. and clinical feedback signals as true label values Spatiotemporal mapping and alignment are performed, and numerical subtraction is executed between the two to quantify the digital residual distribution between the predicted risk and the actual physiological state of the target individual. This extracts the digital bias of risk assessment, i.e., the risk assessment bias value, and combines it with a preset learning rate operator to adjust the initial decision parameter set, including the fusion compensation coefficient and the warning boundary threshold. Perform iterative corrections to generate an adaptive decision parameter set. The preset learning rate operator is a scalar constant used to control the evolution rate of parameters, and its value is determined based on the baseline offset vector. The product of the modulus and the preset smoothing coefficient is used to prevent the early warning logic from oscillating due to a single physiological feature mutation, and to ensure the stability of model convergence; the value of the preset smoothing coefficient is derived from the historical signal-to-noise ratio stability index of the radar echo signal.

[0132] The initial decision parameter set This is the initial decision parameter set before the update, serving as the initial weight benchmark for performing incremental calculations; the adaptive decision parameter set It consists of the calibrated first, second, and third fusion compensation coefficients, as well as the first and second early warning boundary thresholds, and is used to perform digital risk assessment for subsequent monitoring cycles.

[0133] The specific update logic of the calibration and early warning decision logic weights of this incremental learning engine is described as follows: Adaptive decision parameter set equal to the initial decision parameter set The sum of the calibration increment term; wherein, the calibration increment term is the product of the preset learning rate operator, the risk assessment bias value, and the feature offset gradient operator; the feature offset gradient operator is obtained by applying the baseline offset vector. The direction vector obtained by performing partial derivative operations with respect to the decision parameter space consisting of the first, second, and third fusion compensation coefficients and the first and second early warning boundary thresholds is used to determine the parameter search path in the adaptive calibration process.

[0134] Example 2:

[0135] This embodiment, based on Embodiment 1, provides a cardiovascular early warning system based on the decoupling of millimeter-wave radar and physiological signals, such as... Figure 4 As shown, the system includes a regional signal decoupling module, a physiological feature extraction module, a risk linkage module, an intelligent early warning decision-making module, and an adaptive calibration module;

[0136] The regional signal decoupling module is used to determine the origin of the acquired raw radar data stream. Obtain focused echo flow And based on the spatiotemporal quadratic mapping function, the focused echo flow is... Transformed into a cavity containing a thoracic passage Abdominal passage and peripheral channels Regional synchronization signal set .

[0137] The physiological feature extraction module is used to extract regional synchronization signal sets. Mapped to a set of modal components And input it into a deep residual network to generate a heartbeat interval that is included in each beat. Atrial fibrillation probability score and pulse wave conduction time Physiological feature vector Based on regional synchronization signal set Fluid retention index was obtained from the abdominal and thoracic phase data sequences. .

[0138] The risk linkage module is used to base its actions on physiological feature vectors. and fluid retention index Generate a linkage feature matrix by executing the heart failure-stroke correlation mapping logic. Based on the linkage feature matrix Execute dangerous pattern recognition logic to generate interaction enhancement coefficients .

[0139] The intelligent early warning decision module is used to determine the physiological feature vector. and linkage feature matrix Execute a dual-path risk assessment logic to generate independent risk score pairs, combined with an interaction enhancement coefficient. Implement digital integration logic to obtain a comprehensive risk assessment value. And using segmented threshold discrimination logic to determine the comprehensive risk assessment value Mapped to tiered early warning decision instructions .

[0140] The adaptive calibration module is based on a comprehensive risk assessment value. Generate baseline offset vector Combined with clinical feedback signals obtained from medical interactive terminals For the initial decision parameter set including fusion compensation coefficient and early warning boundary threshold Execute incremental adaptive calibration logic and output an adaptive decision parameter set. .

[0141] The parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0142] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling, characterized in that, include: Based on the collected raw radar data stream, a focused echo stream is obtained, and based on the spatiotemporal quadratic mapping function, the focused echo stream is transformed into a regional synchronization signal set containing the thoracic cavity channel, abdominal channel, and peripheral limb channel. The regional synchronization signal set is mapped to a modal component set and input into a deep residual network to generate a physiological feature vector containing beat-by-beat heart rate interval, atrial fibrillation probability score and pulse wave conduction time. The fluid retention index is obtained based on the abdominal and chest phase data sequences of the regional synchronization signal set. Based on the physiological feature vector and fluid retention index, the heart failure-cerebral infarction correlation mapping logic is executed to generate a linkage feature matrix, and the risk pattern recognition logic is executed based on the linkage feature matrix to generate the interaction enhancement coefficient; The dual-path risk assessment logic is executed based on physiological feature vectors and linkage feature matrices to generate independent risk score pairs. The digital fusion logic is then combined with the interaction enhancement coefficient to obtain a comprehensive risk assessment value. Finally, the comprehensive risk assessment value is mapped to a graded early warning decision instruction using segmented threshold discrimination logic. Based on the comprehensive risk assessment value, a baseline offset vector is generated. Combined with the clinical feedback signal obtained from the medical interactive terminal, incremental adaptive calibration logic is executed on the initial decision parameter set, which includes the fusion compensation coefficient and the early warning boundary threshold, and an adaptive decision parameter set is output.

2. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The method for generating the regional synchronization signal set includes: An adaptive spatial beamforming operator based on spatial topology compensation is constructed to perform complex weighted summation on the acquired raw radar data stream to generate a focused echo stream for a specific anatomical region; A logical judgment matrix is ​​constructed based on the collected target distance and azimuth angle. The signal components falling within the corresponding spatial thresholds of the chest, abdomen and limbs are extracted from the focused echo stream through a spatiotemporal quadratic mapping function. The signals are then reconstructed to obtain a regional synchronous signal set composed of the chest cavity channel, the abdominal channel and the limb peripheral channel.

3. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The method for obtaining the physiological feature vector includes: The regional synchronization signal set is mapped to the sum of a finite number of narrowband mode functions using variational mode decomposition operators. The global minimum is then performed to obtain the modal component set, with the goal of minimizing the sum of the estimated bandwidths of each narrowband mode function. A deep residual network is used to map the modal components of the chest cavity to the beat-by-beat heart rate interval, and the modal components of the chest cavity, abdomen and peripheral limbs to the atrial fibrillation probability score. The spatiotemporal cross-correlation peak time delay of the chest cavity channel and the peripheral limb channel is mapped to the pulse wave conduction time, generating a physiological feature vector.

4. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The method for generating the fluid retention index includes: during the monitoring period, retrieving the abdominal phase data sequence of the abdominal channel and the chest phase data sequence of the thoracic channel, performing time integration on the absolute value of the difference between the abdominal phase data sequence and the chest phase data sequence, and generating the fluid retention index by combining a preset attenuation compensation factor and a static reflection reference intensity.

5. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The method for generating the interaction enhancement coefficient includes: Based on the physiological feature vector and fluid retention index, the heart failure-cerebral infarction correlation mapping logic is executed to generate a linkage feature matrix containing covariance components, vascular resistance evolution components, arrhythmia components, and nonlinear enhancement components. Based on the linkage feature matrix, the dangerous pattern recognition logic is executed. The risk probability mapping value is generated by combining the normalized exponential function. The risk probability mapping value is multiplied by a nonlinear enhancement exponent raised to the power of a preset risk coupling constant to obtain the interaction enhancement coefficient.

6. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The execution method of the heart failure-cerebral infarction correlation mapping logic includes: Construct a linkage feature matrix, and perform component mapping on the linkage feature matrix using physiological feature vectors and fluid retention index. Specifically, this includes generating covariance components using the covariance operators of atrial fibrillation probability score and fluid retention index during the monitoring period, and using the covariance components as the first row and first column elements of the linkage feature matrix. Calculate the first derivative of the pulse wave conduction time with respect to the sampling time, and use the product of the first derivative and the preset first weighting coefficient as the vascular resistance evolution component. Map the vascular resistance evolution component to the first row and second column element of the linkage feature matrix. The heart rate interval variation of the heart rate interval under adjacent sampling periods is obtained, and the product of the reciprocal of the heart rate interval variation and the preset second weighting coefficient is used as the arrhythmia component. The arrhythmia component is mapped to the second row and first column element of the linkage feature matrix. The implicit synergistic features between the physiological feature vector and the fluid retention index are extracted using the residual correlation operator. The product of the implicit synergistic features and the preset third weighting coefficient is used as a nonlinear enhancement component. The nonlinear enhancement component is mapped to the second row and second column element of the linkage feature matrix. The values ​​of the preset first weighting coefficient, the preset second weighting coefficient, and the preset third weighting coefficient are preset values ​​based on the basic medical history record of the target individual.

7. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The mapping method for the hierarchical early warning decision instructions includes: Based on the linkage feature matrix and physiological feature vector, a dual-path risk assessment logic is executed to generate independent risk score pairs containing heart failure risk index and cerebral infarction risk index; A digital fusion logic is performed on the independent risk score pairs and the interaction enhancement coefficient. The product of the heart failure risk index and the preset first fusion compensation coefficient, the product of the cerebral infarction risk index and the preset second fusion compensation coefficient, and the product of the interaction enhancement coefficient and the preset third fusion compensation coefficient are summed to generate a comprehensive risk assessment value. Based on the comprehensive risk assessment value, the segmented threshold discrimination logic is executed. Combined with the preset early warning boundary threshold, the risk judgment interval where the comprehensive risk assessment value is located is determined, and the hierarchical early warning decision instruction with the status bit assigned to the first-level flag bit, the second-level flag bit, or the third-level flag bit is output.

8. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The execution method of the dual-path risk assessment logic includes: The covariance component and nonlinear enhancement component in the linkage feature matrix are used as dynamic calibration factors to perform real-time weight correction on each feature component in the physiological feature vector. The dual-path risk assessment logic includes a synchronous heart failure risk assessment path and a stroke risk assessment path. In the heart failure risk assessment path, a first set of features consisting of beat-by-beat heart rate interval, fluid retention index, and heart rate variability is transformed by a first nonlinear feature mapping operator and then aggregated and summed with the first feature weight coefficients corrected by the dynamic calibration factor to generate a heart failure risk index. The value of heart rate variability is obtained based on the statistical analysis of the standard deviation of the beat-by-beat heart rate interval within a preset monitoring window. In the stroke risk assessment pathway, the second feature set, consisting of atrial fibrillation probability score, pulse wave transit time, and blood pressure variability, is transformed by the second nonlinear feature mapping operator and then aggregated and summed with the second feature weight coefficients corrected by the dynamic calibration factor to generate a stroke risk index; the value of blood pressure variability is calculated based on the fluctuation dispersion of the pulse wave transit time within a preset time window.

9. The cardiovascular early warning method based on millimeter-wave radar and physiological signal decoupling according to claim 1, characterized in that, The method for outputting the adaptive decision parameter set includes: Based on the timestamp index of the sampling time, the historical data stream of the accumulated comprehensive risk assessment values ​​from multiple sampling periods is sorted into a historical time series. The static physiological baseline of the historical time series within the preset resting period is extracted, and a baseline offset vector is generated based on the instantaneous comprehensive risk assessment value and the static physiological baseline. The risk judgment deviation value is calculated based on the comprehensive risk assessment value and the clinical feedback signal obtained from the medical interactive terminal. The partial derivative operation is performed on the baseline offset vector to obtain the feature offset gradient operator. Combined with the preset learning rate operator, the initial decision parameter set containing the first, second and third fusion compensation coefficients and the warning boundary threshold is incrementally adaptively calibrated, and the adaptive decision parameter set is output.

10. A cardiovascular early warning system based on millimeter-wave radar and decoupling of physiological signals, used to implement the cardiovascular early warning method based on millimeter-wave radar and decoupling of physiological signals as described in any one of claims 1-9, characterized in that, The system includes a regional signal decoupling module, a physiological feature extraction module, a risk linkage module, an intelligent early warning decision-making module, and an adaptive calibration module; The regional signal decoupling module is used to obtain a focused echo stream based on the acquired raw radar data stream, and to convert the focused echo stream into a regional synchronization signal set containing the thoracic cavity channel, abdominal channel and peripheral limb channel according to the spatiotemporal domain quadratic mapping function. The physiological feature extraction module is used to map the regional synchronization signal set into a modal component set and input it into a deep residual network to generate a physiological feature vector containing beat-by-beat heart rate interval, atrial fibrillation probability score and pulse wave conduction time, and to obtain the fluid retention index based on the abdominal and chest phase data sequences of the regional synchronization signal set. The risk linkage module is used to generate a linkage feature matrix by performing heart failure-cerebral infarction correlation mapping logic based on physiological feature vectors and fluid retention index, and to perform hazard pattern recognition logic based on the linkage feature matrix to generate interaction enhancement coefficients. The intelligent early warning decision module is used to execute dual-path risk assessment logic based on physiological feature vectors and linkage feature matrices to generate independent risk score pairs, combine interaction enhancement coefficients to execute digital fusion logic to obtain a comprehensive risk evaluation value, and use segmented threshold discrimination logic to map the comprehensive risk evaluation value into a graded early warning decision instruction. The adaptive calibration module generates a baseline offset vector based on the comprehensive risk assessment value, combines it with the clinical feedback signal obtained from the medical interactive terminal, performs incremental adaptive calibration logic on the initial decision parameter set including the fusion compensation coefficient and the early warning boundary threshold, and outputs an adaptive decision parameter set.

Images