A web-based big data driven stroke risk prediction method

Abstract

The present application relates to the technical field of medical prediction, in particular to a web-based big data-driven stroke risk prediction method, comprising the following steps: S1, multi-scale data acquisition and alignment: physiological data, behavior data and health status data of a user are collected; S2, causal path identification: a causal feature subgraph with a confidence rating is generated; S3, feature fusion modeling: fluctuation features in real-time physiological data are modeled in response, trend features are extracted from behavior data, and fusion weights of fluctuation features and trend features are dynamically allocated to generate unified risk representation features; S4, risk evolution simulation: the influence of different intervention strategies on stroke risk is simulated to generate a risk evolution trajectory; S5, intervention scheme generation: a personalized intervention scheme is generated. The present application not only improves the accuracy of risk prediction, but also can respond to changes in the patient's health status in real time.

Description

Technical Field

[0001] This invention relates to the field of medical prediction technology, and in particular to a web-based big data-driven method for predicting stroke risk. Background Technology

[0002] Stroke, as one of the leading causes of death and disability worldwide, has a serious impact on patients' health and quality of life. With the increasing aging of the population, the incidence of stroke continues to rise. Early prediction and intervention have become the key to effectively reducing stroke mortality and disability rates. Traditional stroke prediction methods usually rely on the subjective assessment of patients' symptoms and signs by clinicians, combined with routine health check data, such as blood pressure and blood sugar. These methods can predict the occurrence of stroke to a certain extent, but due to the lack of accurate real-time data support and personalized intervention strategies, their predictive accuracy and treatment effectiveness are still quite limited.

[0003] Currently, most stroke prediction methods rely on a single data source or regular medical examinations. These traditional methods face the following problems: First, they lack real-time collection and analysis of multi-dimensional data, failing to fully reflect the dynamic changes in patients' health status. Second, existing methods typically lack causal relationship analysis, failing to delve into the interactions between different health factors, resulting in an inability to provide accurate personalized intervention recommendations. Third, they lack real-time monitoring and intelligent feedback mechanisms for patients' health status, making it difficult to intervene in a timely and effective manner based on changing risks. These shortcomings limit the effectiveness of existing technologies in practical applications, particularly in their limited capabilities for personalized, dynamic prediction and intervention. Summary of the Invention

[0004] This invention provides a web-based big data-driven stroke risk prediction method that responds in real time to changes in a patient's health, improves the accuracy of stroke prediction and treatment effectiveness, and reduces the incidence and disability rate of stroke.

[0005] A web-based, big data-driven stroke risk prediction method includes the following steps:

[0006] S1, Multi-scale data acquisition and alignment: Collect users' physiological data, behavioral data and health status data, and achieve multi-scale data alignment across time scales through quantum timestamps;

[0007] S2, Causal Path Identification: Aligned multi-scale data is input into the causal discovery engine to identify the causal relationship between physiological abnormalities and health factors, and to generate a causal feature subgraph with confidence rating.

[0008] S3, Feature Fusion Modeling: Based on the causal feature subgraph, a time-varying weighted network is constructed. The spiking neural network (SNN) is used to model the response of fluctuation features in real-time physiological data. The temporal convolutional network (TCN) is combined to extract trend features from behavioral data. The fusion weights of fluctuation features and trend features are dynamically allocated according to the user's current health status data to generate a unified risk characterization feature.

[0009] S4, Risk Evolution Simulation: Input the generated risk representation features into a causal reinforcement learning framework to simulate the impact of different intervention strategies on stroke risk and generate risk evolution trajectories;

[0010] S5, Intervention Plan Generation: Based on the results of risk evolution simulation, a three-dimensional early warning display is triggered, and a personalized intervention plan is generated.

[0011] Optionally, the multi-scale data acquisition and alignment in S1 includes:

[0012] S11, Multi-scale data classification and collection: Collect users' physiological data, behavioral data and health status data. Physiological data includes cerebral blood flow, ECG RR interval and blood pressure waveform. Behavioral data includes motion acceleration, angular velocity and voice characteristics. Health status data comes from the electronic medical record system and includes stroke score, disability assessment and medication adherence indicators.

[0013] S12, Quantum Timestamp Generation and Binding: Obtain a time reference through a quantum key distribution network and attach a unique quantum timestamp to each type of data;

[0014] S13, Cross-scale time alignment: Millisecond-level compensation for physiological data, periodic correction for behavioral data, and alignment of health status data according to calendar event mapping.

[0015] S14, Timestamp Verification and Anomaly Handling: Time difference detection is performed on the collected multi-scale data, and the validity of the timestamp is verified by quantum signature to remove non-compliant data.

[0016] Optionally, the cross-scale time alignment in S13 includes:

[0017] S131, Physiological data alignment: The physiological data acquisition delay is corrected by a linear compensation model, and the original time is aligned to the quantum time reference by the device calibration coefficient to achieve millisecond-level synchronization;

[0018] S132, Behavioral data alignment: For periodic time errors in behavioral data, a sine function correction method is used to perform second-level alignment to eliminate time deviations caused by environmental interference and device clock offset;

[0019] S133, Health Status Data Alignment: Align health status data by calendar time segments and correct it by time slice mapping and medical system working time offset.

[0020] Optionally, the causal path identification in S2 includes:

[0021] S21, Causal Feature Space Construction: The output aligned multi-scale data is standardized and a three-dimensional causal feature tensor is constructed.

[0022] S22, Hybrid Causal Structure Learning: Discovering static causal relationships between variables through constrained Bayesian networks and verifying time-series causality between variables through temporal Granger causality tests;

[0023] S23, Elimination of False Associations: Through Counterfactual Causality Strength The analysis screened out weakly correlated or spuriously correlated paths, retaining only those that met the criteria. The causal edges are established, and environmental variable control tests are performed simultaneously.

[0024] S24, Causal Subgraph Generation: Based on the time lag range of causal edges, a multi-layer causal network structure is constructed, specifically including:

[0025] First layer: Physiology Physiological causal edge, delay ;

[0026] Second level: Behavior Physiological causal boundary, ;

[0027] Third level: Health status Behavioral / physiological causal boundary ;

[0028] The weights of each causal edge in the multi-layer causal network structure are calculated jointly based on normalized causal strength and Granger test strength.

[0029] S25, Confidence Rating: A confidence score is generated by integrating counterfactual causality strength, temporal Granger causality test results, and causal structure stability. And graded assessment, when At that time, the confidence level was A. At that time, the confidence level was B. At that time, the confidence level was C (manual review required).

[0030] Optionally, the hybrid causal structure learning in S22 includes:

[0031] S221, Static Causal Discovery: A constrained Bayesian network is used for causal graph structure learning. Conditional mutual information (MI) is calculated to identify causal relationships between variables, and sparsity control coefficients are utilized. Optimize;

[0032] S222, Dynamic Causality Verification: Granger causality test is used to verify causal paths, and the F-test statistic is calculated. And filter out paths with temporal causal relationships.

[0033] Optionally, the feature fusion modeling in S3 includes:

[0034] S31, Parallel Extraction of Multimodal Features: A dual-channel architecture is used to extract multimodal features in parallel. Through the Spike Neural Network (SNN) channel, a Leaky Integral Ignition (LIF) neuron model is used to process real-time fluctuation features. Through the Temporal Convolutional Network (TCN) channel, a 6-layer residual dilated convolutional network is used to extract trend features.

[0035] S32, Dynamic Weight Adaptive Fusion: Extract the health status vector from the health status data, use a multilayer perceptron (MLP) to generate dynamic fusion weights from the outputs of the spiking neural network (SNN) channel and the temporal convolutional network (TCN) channel, and combine the calculated dynamic fusion weights to generate the final risk representation.

[0036] Optionally, the parallel extraction of multimodal features in S31 includes:

[0037] S311, Real-time Fluctuation Modeling: Synaptic connections are initialized based on the edge weights of the causal subgraph, and the membrane potential is updated through the Leakage Integral Ignition (LIF) neuron model. By setting an adaptive threshold, it is determined whether the neuron fires a pulse to capture fluctuation characteristics.

[0038] S312, Trend Feature Extraction: By configuring a 6-layer residual dilated convolutional network, each layer uses dilated convolution operations to extract trend information at different scales, and by using a skip connection mechanism, the output of each convolution layer is combined with the input to enhance the feature representation capability.

[0039] Optionally, the dynamic weight adaptive fusion in S32 includes:

[0040] S321, Spatiotemporal coding of health status: Features of health status data are extracted through multi-scale convolution, information at different time scales is captured by dilated convolution, and a health status vector is generated by pooling to extract periodic health patterns.

[0041] S322, Gated Weight Generation: Dynamic fusion weights are generated from the outputs of the spiking neural network (SNN) and temporal convolutional network (TCN) channels using a multilayer perceptron (MLP). The weights are calculated by combining the health state vector with gating operations and the weights are adjusted using a time period modulation factor.

[0042] S323, Risk Characterization Synthesis: The output feature maps of the spiking neural network (SNN) channel and the temporal convolutional network (TCN) channel are weighted and fused according to the adjusted dynamic weights to generate the final risk characterization features.

[0043] Optionally, the risk evolution simulation in S4 includes:

[0044] S41, Virtual Patient Environment Construction: The generated final risk characterization features are mapped to the environmental state of the virtual patient. The high-dimensional features are mapped to a 512-dimensional space through dimensionality reduction by an autoencoder. Constraints are set for multi-dimensional intervention measures through a multi-dimensional intervention strategy space, including drug dosage, rehabilitation intensity and diet plan.

[0045] S42, Causal Transition Function Modeling: Construct a neural differential equation (NDE) model to describe the causal relationship of state transitions in a virtual patient environment. By defining state transition equations, simulating random physiological fluctuations and combining them with causal transition functions, analyze the changes in health status over time. And through a compound reward function, evaluate the effectiveness of each intervention, including the amount of risk change, intervention cost, and compliance.

[0046] S43, Strategy Optimization and Simulation: A causal-guided strategy gradient algorithm is used to generate behavioral strategies, and multiple instances are run in parallel simulation to optimize the intervention strategy and update the patient's health status until the simulation time reaches the set number of months.

[0047] S44, Risk Trajectory Generation: Kernel density estimation is performed on the simulation results to analyze the evolution trajectory of the risk. By calculating the risk distribution at each time point and extracting risk threshold crossing events, key risk change moments are marked.

[0048] Optionally, the intervention plan generation in S5 includes:

[0049] S51, Analyze Risk Changes: Based on the results of risk evolution simulation, extract the amount of risk change and the key moments of risk change;

[0050] S52, Trigger 3D Early Warning Display: When a risk change exceeds 0.2 or a critical risk change moment is detected, a 3D early warning display is automatically triggered, which displays the patient's risk status in real time through 3D visualization graphics;

[0051] S53, Generate Personalized Intervention Plans: Based on the results of the 3D early warning display, combined with the patient's personalized information (such as age, gender, medical history, etc.), generate personalized intervention plans, including drug treatment, rehabilitation training, and dietary adjustments.

[0052] The beneficial effects of this invention are:

[0053] This invention acquires multimodal data, including physiological data, behavioral data, and health status data, and uses quantum timestamp technology to precisely align multi-source data across time scales, ensuring high data quality and timeliness. This enables synchronous processing of various types of data, ensuring data comparability and usability.

[0054] This invention, through causal path identification and feature fusion modeling, can identify causal relationships between different health factors and generate unified risk characterization features. By combining spiking neural networks and temporal convolutional networks to dynamically fuse multiple features, it not only improves the accuracy of risk prediction but also responds in real time to changes in patients' health status, providing strong support for clinical decision-making.

[0055] This invention, by combining drug therapy, rehabilitation training, and dietary adjustments, can effectively reduce the risk of stroke in patients and improve treatment outcomes. The combination of three-dimensional early warning display and personalized intervention plans can provide medical personnel with real-time and intuitive risk assessment and intervention suggestions, ensuring timely and accurate treatment. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in this 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 for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of the prediction method flow according to an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of causal path identification according to an embodiment of the present invention. Detailed Implementation

[0059] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0060] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0061] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0062] like Figures 1-2 As shown, a web-based big data-driven stroke risk prediction method includes the following steps:

[0063] S1, Multi-scale data acquisition and alignment: Collect users' physiological data, behavioral data and health status data, and achieve multi-scale data alignment across time scales through quantum timestamps;

[0064] S2, Causal Path Identification: Aligned multi-scale data is input into the causal discovery engine to identify the causal relationship between physiological abnormalities and health factors, and to generate a causal feature subgraph with confidence rating.

[0065] S3, Feature Fusion Modeling: Based on the causal feature subgraph, a time-varying weighted network is constructed. The spiking neural network (SNN) is used to model the response of fluctuation features in real-time physiological data. The temporal convolutional network (TCN) is combined to extract trend features from behavioral data. The fusion weights of fluctuation features and trend features are dynamically allocated according to the user's current health status data to generate a unified risk characterization feature.

[0066] S4, Risk Evolution Simulation: Input the generated risk representation features into a causal reinforcement learning framework to simulate the impact of different intervention strategies on stroke risk and generate risk evolution trajectories;

[0067] S5, Intervention Plan Generation: Based on the results of risk evolution simulation, a three-dimensional early warning display is triggered, and a personalized intervention plan is generated.

[0068] Multi-scale data acquisition and alignment in S1 includes:

[0069] S11, Multi-scale Data Classification and Acquisition: Collects users' physiological data, behavioral data, and health status data. Physiological data includes cerebral blood flow, ECG RR interval, and blood pressure waveforms; behavioral data includes motion acceleration, angular velocity, and voice characteristics; health status data comes from the electronic medical record system and includes stroke scores, disability assessments, and medication adherence indicators, specifically including:

[0070] Physiological data acquisition: Physiological data is acquired using a medical-grade wearable device at a sampling rate of 1000Hz;

[0071] Behavioral data acquisition: Limb motion acceleration was acquired using a nine-axis IMU sensor at a sampling rate of 100Hz. angular velocity And speech features, wherein the speech features are extracted using the Mel-frequency cepstral coefficient (MFCC) method, and are expressed as:

[0072] ;

[0073] in, For discrete cosine transform, For the first Mel frequency cepstral coefficients, For Mel filter banks, This represents the power spectrum of the speech frame. For Fast Fourier Transform;

[0074] Health status data collection: Extracting health status data from the electronic health record (EHR) system, specifically including:

[0075] Stroke score: ;

[0076] Disability assessment: ;

[0077] Medication adherence indicators: ;

[0078] S12, Quantum Timestamp Generation and Binding: A time reference is obtained through a quantum key distribution network, and a unique quantum timestamp is attached to each type of data, represented as:

[0079] ;

[0080] in, The initial deviation between the device and the quantum clock source. Duration of the synchronization operation The time synchronization decay constant, For quantum timestamps, To coordinate world time;

[0081] ;

[0082] in, The final generated timestamp includes both time and hash. The hash value of the data block;

[0083] S13, Cross-scale time alignment: Millisecond-level compensation is applied to physiological data, periodic correction is applied to behavioral data, and health status data is aligned according to calendar event mapping to ensure time consistency of multi-source data;

[0084] S14, Timestamp Verification and Anomaly Handling: Time difference detection is performed on the collected multi-scale data, and the validity of the timestamps is verified through quantum signatures to eliminate non-compliant data, ensuring data quality and synchronization accuracy. This is represented as:

[0085] ;

[0086] in, The maximum allowable time interval deviation, (Physiological data) (Behavioral data) (Health data) , The time interval between two consecutive data collections;

[0087] ;

[0088] in, To verify the results, use a boolean type (True / False). This is a quantum key-based signature verification algorithm. This is the public key.

[0089] Cross-scale time alignment in S13 includes:

[0090] S131, Physiological Data Alignment: A linear compensation model is used to correct the physiological data acquisition delay. The original time is aligned to a quantum time reference using equipment calibration coefficients, achieving millisecond-level synchronization, as shown below:

[0091] ;

[0092] in, This represents the time delay coefficient of the biosensor. The corrected timestamps for physiological data. The original data recording time;

[0093] S132, Behavioral Data Alignment: For periodic time errors in behavioral data, a sine function correction method is used for second-level alignment to eliminate time deviations caused by environmental interference and device clock skew. This is expressed as:

[0094] ;

[0095] in, Environmental disturbance intensity factor For the sensor crystal oscillator frequency, The corrected timestamp for the behavioral data;

[0096] S133, Health Status Data Alignment: Align health status data by calendar time segments, corrected by time slice mapping and medical system operating time offset, ensuring a unified time reference on a daily scale, represented as:

[0097] ;

[0098] in, The length of a single day's time slice. This is due to a deviation from the standard working hours of medical institutions. This is the timestamp for the aligned health data.

[0099] Causal path identification in S2 includes:

[0100] S21, Causal Feature Space Construction: The output aligned multi-scale data is standardized, and a three-dimensional causal feature tensor is constructed, represented as:

[0101] ;

[0102] in, For multi-scale data In the sliding window The average value over the seconds. For multi-scale data In the statistical period The dynamic standard deviation within hours, This refers to multi-scale data after standardization.

[0103] ;

[0104] in, For a three-dimensional causal feature tensor, A set of physiological characteristic indexes, A set of behavioral feature indexes A set of health status feature indexes. , , These are the corresponding standardized eigenvalues;

[0105] S22, Hybrid Causal Structure Learning: Discovering static causal relationships between variables through constrained Bayesian networks and verifying the temporal causality between variables through temporal Granger causality tests, thereby improving the temporal explanatory power of causal structures;

[0106] S23, Elimination of False Associations: Through Counterfactual Causality Strength The analysis screened out weakly correlated or spuriously correlated paths, retaining only those that met the criteria. The causal edges are identified, and environmental variable control tests are performed simultaneously to retain the core causal links with practical intervention value, represented as:

[0107] ;

[0108] ;

[0109] in, For causal variables in causal path analysis, As the target variable in causal analysis, This refers to confounding variables used for control or regulation. , Causal variables Two different states, Pearl's intervention symbol. For the set of all observed variables, This represents a subset of confounding variables used to test for conditional independence. express and exist Independent under certain conditions;

[0110] S24, Causal Subgraph Generation: Based on the time lag range of causal edges, a multi-layer causal network structure is constructed, specifically including:

[0111] First layer: Physiology Physiological causal edge, delay ;

[0112] Second level: Behavior Physiological causal boundary, ;

[0113] Third level: Health status Behavioral / physiological causal boundary ;

[0114] The weights of each causal edge in a multi-layered causal network structure are calculated jointly using normalized causal strength and Granger test strength, and are expressed as follows:

[0115] ;

[0116] in, The weight of the causal edge. For the first The strength of a causal edge, The maximum causal strength among all edges. This is the F-statistic for Granger causality testing;

[0117] S25, Confidence Rating: A confidence score is generated by integrating counterfactual causality strength, temporal Granger causality test results, and causal structure stability. And graded assessment, when At that time, the confidence level was A. At that time, the confidence level was B. At that time, the confidence level is C (requires manual review), which is expressed as:

[0118] ;

[0119] in, These are the corresponding weighting coefficients. Parent node of the target node structural entropy, parent node The probability distribution, parent node The log probability of each variable taking a value.

[0120] Hybrid causal structure learning in S22 includes:

[0121] S221, Static Causal Discovery: A constrained Bayesian network is used for causal graph structure learning. Conditional mutual information (MI) is calculated to identify causal relationships between variables, and sparsity control coefficients are utilized. Optimize by retaining the most important causal paths, represented as:

[0122] ;

[0123] in, It is a Bayesian causal graph structure. For the input dataset, The posterior probability is used to learn the optimal graph structure. For variables right There is a causal relationship. For variables The set of parent nodes, For conditional mutual information, measure the condition in a given context. In this case, and The degree of correlation;

[0124] S222, Dynamic Causality Verification: Granger causality test is used to verify causal paths, and the F-test statistic is calculated. And filter out the paths with temporal causal relationships, represented as:

[0125] ;

[0126] in, For the sum of squared residuals of the constrained model (excluding) ), The sum of squared residuals of the unconstrained model (including ), This represents the lag order, corresponding to three time granularities (seconds, minutes, and hours). To be the effective sample size, the following conditions must be met. ;

[0127] The criteria for selecting paths with temporal causal relationships are as follows:

[0128] ;

[0129] in, The critical value of the F-distribution, confidence level Degrees of freedom .

[0130] Feature fusion modeling in S3 includes:

[0131] S31, Parallel Extraction of Multimodal Features: A dual-channel architecture is used to extract multimodal features in parallel. Through the Spike Neural Network (SNN) channel, a Leaky Integral Ignition (LIF) neuron model is used to process real-time fluctuation features. Through the Temporal Convolutional Network (TCN) channel, a 6-layer residual dilated convolutional network is used to extract trend features.

[0132] S32, Dynamic Weight Adaptive Fusion: Extract the health status vector from the health status data, use a multilayer perceptron (MLP) to generate dynamic fusion weights from the outputs of the spiking neural network (SNN) channel and the temporal convolutional network (TCN) channel, and combine the calculated dynamic fusion weights to generate the final risk representation.

[0133] Parallel extraction of multimodal features in S31 includes:

[0134] S311, Real-time Fluctuation Modeling: Synaptic connections are initialized based on causal subgraph edge weights, and membrane potentials are updated using a Leakage Integral Fire (LIF) neuron model. An adaptive threshold is set to determine whether a neuron fires a pulse, capturing fluctuation characteristics. Specifically, this includes:

[0135] S3111, LIF neuron initialization: Synaptic connections are configured based on the edge weights of the causal subgraph, represented as:

[0136] ;

[0137] in, For the first The first neuron and the second Synaptic connection weights between neurons Let the causal edge weights come from the causal subgraph. , representing the randomness of connections in each neuron, follows a uniform distribution. ;

[0138] S3112, Dynamic Update of Membrane Potential: The leaky integral ignition (LIF) neuron model is used to dynamically update the neuronal membrane potential, expressed as:

[0139] ;

[0140] in, For the first The membrane potential of a neuron The membrane time constant is used to match the cerebral vascular pressure fluctuation cycle. This is the leakage compensation coefficient, representing the degree of membrane potential decay. The input current represents the current generated by the activity of the presynaptic neuron. This is the output signal of the presynaptic neuron;

[0141] S3113, pulse triggering mechanism: ;

[0142] in, For the first The pulse output state of each neuron, where 1 indicates firing a pulse and 0 indicates not firing a pulse. As an adaptive threshold, a pulse is fired when the neuron's membrane potential reaches this value;

[0143] S312, Trend Feature Extraction: A 6-layer residual dilated convolutional network is configured, with each layer using dilated convolution operations to extract trend information at different scales. The convolution operations ensure feature dimension matching, and a skip connection mechanism combines the output and input of each convolutional layer to enhance feature representation capabilities. Specifically, this includes:

[0144] S3121, Residual Block Structure Configuration: (The following is a continuation of the previous sentence) layer The parameter configuration includes the expansion rate. kernel size Number of output channels ;

[0145] The input-output dimension matching representation of a convolution operation is as follows:

[0146] ;

[0147] in, The time length (or length dimension) of the output feature map. The time length (or length dimension) of the input feature map. The padding length is used to ensure boundary alignment during convolution operations. The stride of the convolution operation controls the step size of the convolution kernel's movement.

[0148] S3122, Dilated Convolution Calculation: ;

[0149] in, For the first The output feature map of the layer, For the first The convolutional kernel weights of the layer, For the first The output feature map of the layer, For the first Layer bias terms;

[0150] S3123, Jump Connection Mechanism: ;

[0151] in, For the first Weighted output feature map of skip connections in the layer This is a 1×1 convolution operation used to adjust the number of channels. For the first Input feature map of the layer.

[0152] Dynamic weight adaptive fusion in S32 includes:

[0153] S321, Spatiotemporal Coding of Health Status: This method extracts features from health status data through multi-scale convolution, captures information at different time scales using dilated convolution, and generates health status vectors through pooling to extract periodic health patterns. Specifically, it includes:

[0154] S3211, Multi-scale Convolutional Feature Extraction: Input Health Status Data (in Hour, (3D features), which are encoded through a three-level dilated convolution, and represented as:

[0155] Expansion rate Output dimensions: ;

[0156] Expansion rate Output dimensions: ;

[0157] Expansion rate Output dimensions: ;

[0158] in, The size of the feature encoding convolution kernel, For feature coding expansion rate, Number of output channels For gated linear units, , It is the sigmoid activation function;

[0159] S3212, Temporal Dependency Modeling: A convolutional neural network (CNN) is used to capture the periodic patterns of health status data, represented as:

[0160] Daily cycle pattern extraction;

[0161] Three-day cycle pattern extraction;

[0162] ;

[0163] in, The features of the daily cycle pattern are extracted through convolution operations. The three-day periodic pattern features are extracted through convolution operations. To The feature map obtained after layer normalization and GeLU activation This refers to the kernel size of the convolutional neural network. This represents the dilation rate of the convolutional neural network.

[0164] ;

[0165] in, For the final health state vector, To adapt to max pooling, the dimension of the feature map is compressed to a fixed size (512).

[0166] S322, Gated Weight Generation: Dynamically fused weights are generated using a multilayer perceptron (MLP) to combine the outputs of the spiking neural network (SNN) and temporal convolutional network (TCN) channels. Weights are calculated through gating operations combined with a health state vector, and the weights are adjusted using a time-period modulation factor. Specifically, this includes:

[0167] S3221, Multilayer Perceptron Architecture: Constructs a gated projection network to generate dynamic weights, represented as:

[0168] ;

[0169] ;

[0170] ;

[0171] in, The dynamic weights output by the gating mechanism. Here is the weight matrix of the gated network. For the bias term of the gated network, Features of the projection network output Here is the weight matrix of the projection network. For the bias term of the projection network, For the dynamic weights of the SNN channels, The dynamic weights of the TCN channel;

[0172] S3222, Phase-Sensitive Adjustment: A time-period modulation factor is introduced to dynamically adjust the weights, expressed as:

[0173] ;

[0174] ;

[0175] in, , These are the adjusted dynamic weights of the SNN channels and the TCN channels, respectively. As a moderating factor, it controls the impact of changes in health status on the weights. This is the current timestamp;

[0176] S323, Risk Characteristic Synthesis: The output feature maps of the spiking neural network (SNN) channel and the temporal convolutional network (TCN) channel are weighted and fused according to adjusted dynamic weights to generate the final risk characterization features. The fusion process is ensured to meet predetermined constraints, specifically including:

[0177] S3231, Feature Dimension Alignment: Align the dimensions of the SNN and TCN outputs to ensure a matching number of channels, expressed as:

[0178] ;

[0179] ;

[0180] in, , These are the output feature maps for the SNN and TCN channels, respectively. , The feature map is obtained after linear transformation and is used for fusion. The desired dimension number is obtained after linear transformation of the output feature maps of SNN and TCN;

[0181] S3232, Dynamic Weighted Fusion: Weighted fusion is performed based on adjusted dynamic weights to generate the final risk representation, expressed as:

[0182] ;

[0183] in, The final risk characterization features;

[0184] Constraints: ;

[0185] in, , These are the traces of the fusion weights of the SNN and TCN, respectively, ensuring that their sum is 1.

[0186] Risk evolution simulations in S4 include:

[0187] S41, Virtual Patient Environment Construction: The generated final risk characterization features are mapped to the environmental state of the virtual patient. Dimensionality reduction using an autoencoder maps high-dimensional features to a 512-dimensional space. Constraints are set for multi-dimensional intervention strategies, including drug dosage, rehabilitation intensity, and dietary plans, to ensure effective control during simulation. Specifically, this includes:

[0188] S411, State space definition: Mapping the generated final risk representation features to environmental states, represented as:

[0189] ;

[0190] in, Let the environment state vector be... This is a dimension reduction projection based on an autoencoder;

[0191] S412, Action Space Design: Define a multi-dimensional intervention strategy space, represented as:

[0192] ;

[0193] in, For intervention strategies;

[0194] Constraints: ;

[0195] ;

[0196] S42, Causal Transfer Function Modeling: A neural differential equation (NDE) model is constructed to describe the causal relationships of state transitions in a virtual patient environment. By defining state transition equations, simulating random physiological fluctuations, and combining this with causal transfer functions, changes in health status over time are analyzed. Furthermore, the effectiveness of each intervention is evaluated using a composite reward function, including changes in risk, intervention costs, and adherence. Specifically, this includes:

[0197] S421, State transition equation: Construct a causal model based on neural differential equations, expressed as:

[0198] ;

[0199] in, It is a three-layer multilayer perceptron (MLP) network with the following structure: , For learnable parameters, It is a heteroscedastic noise network. As independent parameters, This is the Wiener process increment, used to simulate random physiological fluctuations;

[0200] S422, Reward Function Design: Calculate the composite reward, expressed as:

[0201] ;

[0202] in, For the change in risk, To standardize drug costs and rehabilitation costs, , For compliance, the predicted value is based on historical compliance rate, ranging from [missing value]. , , , For adjustable hyperparameters, For the reward function at time step The value;

[0203] S43, Strategy Optimization and Simulation: A causal-guided policy gradient algorithm is used to generate behavioral strategies. Multiple instances are run in parallel simulations to optimize the intervention strategy and update the patient's health status until the simulation period reaches the set number of months. Specifically, this includes:

[0204] S431, Behavioral Policy Generation: A causal-guided policy gradient algorithm is used to generate the behavioral policy, represented as follows:

[0205] ;

[0206] in, For a policy network, it represents the state. Select action The probability distribution, Let be the advantage function, representing the advantage of each action in a given state. The parameters of the policy network determine the policy for selecting control actions;

[0207] S432, Virtual Clinical Trial Execution: Parallel Execution Each simulation instance executes the following:

[0208] ;

[0209] ;

[0210] ;

[0211] ;

[0212] in, This refers to the number of parallel simulation instances used to run multiple virtual patient environment instances simultaneously. The time step is set to 1 month, representing the time interval between each simulation step. The noise is Gaussian, simulating random physiological fluctuations at each time step, derived from a standard normal distribution. The mean is 0, and the covariance matrix is ​​the identity matrix. , For the current moment Intervention strategies, Given a state Generate an action The probability distribution, For noise model, These are the parameters of the noise model;

[0213] S44, Risk Trajectory Generation: Kernel density estimation is performed on the simulation results to analyze the evolution trajectory of the risk. By calculating the risk distribution at each time point and extracting risk threshold crossing events, key risk change moments are marked, represented as:

[0214] ;

[0215] in, In time At any given moment, the probability density function of risk. The number of simulation instances, For kernel function, , This is the bandwidth parameter of the kernel function. The risk value at the current moment. For the first A simulation instance at time Risk value at any given moment;

[0216] ;

[0217] in, The time when a critical event occurs, i.e., when the risk value exceeds the clinical warning threshold. The earliest moment of time, In time Risk value at any given moment This is the clinical early warning threshold.

[0218] Intervention protocol generation in S5 includes:

[0219] S51, Analyze Risk Changes: Based on the results of risk evolution simulation, extract the amount of risk change and the key moments of risk change;

[0220] S52, Trigger 3D Early Warning Display: When a risk change exceeds 0.2 or a critical risk change moment is detected, a 3D early warning display is automatically triggered, which displays the patient's risk status in real time through 3D visualization graphics;

[0221] S53, Generate Personalized Intervention Plans: Based on the results of the 3D early warning display, combined with the patient's personalized information (such as age, gender, medical history, etc.), generate personalized intervention plans, including drug treatment, rehabilitation training, and dietary adjustments, to reduce the patient's risks and improve treatment outcomes.

[0222] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0223] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A web-based big data driven stroke risk prediction method, characterized in that, The method comprises the following steps: S1, multi-scale data acquisition and alignment: physiological data, behavior data and health status data of the user are collected, and multi-scale data alignment across time scales is realized through quantum timestamps; Specifically comprising: S11, multi-scale data classification acquisition: physiological data, behavior data and health status data of the user are collected, the physiological data includes brain blood flow, electrocardiogram RR interval and blood pressure waveform, the behavior data includes motion acceleration, angular velocity and voice features, and the health status data comes from an electronic medical record system, including stroke score, disability assessment and medication compliance indicators; S12, quantum timestamp generation and binding: a time reference is obtained through a quantum key distribution network, and a unique quantum timestamp is attached to each type of data; S13, cross-scale time alignment: millisecond-level compensation is performed on the physiological data, periodic correction is applied to the behavior data, and the health status data is aligned in a calendar event mapping manner; S14, timestamp verification and exception handling: time difference detection is performed on the collected multi-scale data, and the effectiveness of the timestamp is verified through quantum signature verification to eliminate non-compliant data; S2, causal path identification: the aligned multi-scale data is input into a causal discovery engine to identify the causal relationship between physiological abnormalities and health factors, and a causal feature subgraph with a confidence rating is generated; S3, feature fusion modeling: a time-varying weight network is constructed based on the causal feature subgraph, a pulse neural network is used to model the fluctuation features in real-time physiological data, a time convolution network is used to extract trend features from behavior data, and the fusion weight of fluctuation features and trend features is dynamically allocated according to the current health status data of the user, generating a unified risk representation feature; S4, risk evolution simulation: the generated risk representation feature is input into a causal reinforcement learning framework to simulate the impact of different intervention strategies on stroke risk, generating a risk evolution trajectory; S5, intervention scheme generation: based on the results of risk evolution simulation, a three-dimensional early warning display is triggered to generate a personalized intervention scheme; The feature fusion modeling in S3 comprises: S31, parallel extraction of multi-modal features: a dual-channel architecture is used to extract multi-modal features in parallel, a leaky integral firing neuron model is used to process real-time fluctuation features through a pulse neural network channel, and a 6-layer residual dilation convolution network is used to extract trend features through a time convolution network channel; Specifically comprising: S311, real-time fluctuation modeling: synapse connections are initialized based on the edge weights of the causal subgraph, the membrane potential is updated through a leaky integral firing neuron model, and an adaptive threshold is set to determine whether the neuron fires a pulse, capturing fluctuation features; S312, trend feature extraction: a 6-layer residual dilation convolution network is configured, each layer uses a dilation convolution operation to extract trend information of different scales, and a jump connection mechanism is used to combine the output of each convolution layer with the input, enhancing the feature representation capability; S32, dynamic weight adaptive fusion: extract the health state vector in the health state data, generate the dynamic fusion weight of the output of the pulse neural network channel and the time convolution network channel using a multilayer perception, and combine the calculated dynamic fusion weight to generate the final risk representation, specifically including: S321, health state space-time coding: features of the health state data are extracted through multi-scale convolution, information of different time scales is captured using dilated convolution, and a health state vector is generated through pooling to extract periodic health patterns; S322, gate weight generation: a multilayer perception is used to generate the dynamic fusion weight of the output of the pulse neural network channel and the time convolution network channel, the weight is calculated by combining the health state vector through gating operation, and the weight is adjusted using a time period modulation factor; S323, risk representation synthesis: the output feature maps of the pulse neural network channel and the time convolution network channel are weighted and fused according to the adjusted dynamic weight to generate the final risk representation feature.

2. The web-based big data driven stroke risk prediction method of claim 1, wherein, The cross-scale time alignment in S13 includes: S131, physiological data alignment: correct the physiological data collection time delay through a linear compensation model, align the original time to the quantum time reference using the device calibration coefficient, and achieve millisecond-level synchronization; S132, behavior data alignment: for the periodic time error in the behavior data, a sinusoidal function correction method is used for second-level alignment to eliminate the time deviation caused by environmental interference and device clock offset; S133, health state data alignment: the health state data is segmented and aligned according to the calendar time, and the time slice mapping is corrected by the time offset of the medical system. 3.The Web-based big data driven stroke risk prediction method of claim 1, wherein, The causal path identification in S2 includes: S21, construction of causal feature space: the output aligned multi-scale data is standardized and a three-dimensional causal feature tensor is constructed; S22, mixed causal structure learning: static causal relationships between variables are discovered through a constraint-based Bayesian network, and time series causality between variables is verified through a Granger causality test; S23, false association elimination: by counterfactual causal strength Analysis to exclude weak or spurious paths, keeping only those that satisfy causal edges, while conducting environmental variable control tests; S24, causal subgraph generation: based on the time lag range of the causal edges, a multi-layer causal network structure is constructed, specifically including: First tier: physiology physiological causal edge, delay ; Second layer: Behavior physiological cause-effect edge, ; Third tier: Health status Behavioral / physiological cause-effect edges, ; The weights of the causal edges in the multi-layer causal network structure are calculated based on the normalized causal strength and the Granger test strength; S25, confidence rating: fuse counterfactual causal strength, temporal Granger causality test results and causal structure stability to generate confidence score and hierarchical evaluation, when the confidence rating is A level, when the confidence rating is B level, when the confidence rating is C level. 4.The Web-based big data-driven stroke risk prediction method of claim 3, wherein, The mixed causal structure learning in S22 includes: S221, static causal discovery: constraint-based Bayesian network is used to learn the structure of causal graph, the causal relationship between variables is identified by calculating conditional mutual information, and the sparsity control coefficient is used optimization; S222, dynamic causal verification: adopt Granger causality test to verify the causal path, calculate F test statistic and screen out the path with time series causal relationship. 5.The Web-based big data driven stroke risk prediction method of claim 1, wherein, The risk evolution simulation in S4 includes: S41, virtual patient environment construction: the generated final risk representation feature is mapped to the environment state of the virtual patient, high-dimensional features are mapped to a 512-dimensional space through an autoencoder dimension reduction, and multi-dimensional intervention strategy space including drug dosage, rehabilitation intensity and diet scheme is used to set constraint conditions for multi-dimensional intervention measures; S42, causal transfer function modeling: a neural differential equation model is constructed to describe the causal relationship of state transfer in the virtual patient environment, a state transfer equation is defined to simulate random physiological fluctuations, and the change of health state over time is analyzed by combining the causal transfer function, and the effect of each intervention measure is evaluated through a composite reward function, including risk change, intervention cost and compliance; S43, policy optimization and simulation: generate behavior policy using causal-guided policy gradient algorithm, optimize intervention policy by running multiple instances in parallel simulation, update patient's health status until simulation time reaches the set month; S44, risk trajectory generation: perform kernel density estimation on simulation results, analyze the evolution trajectory of risk, calculate the risk distribution at each time point, and extract the risk threshold crossing events to mark the key risk change moments. 6.The Web-based big data-driven stroke risk prediction method of claim 5, wherein, The intervention scheme generation in S5 includes: S51, analyze risk changes: according to the results of risk evolution simulation, extract the risk change amount and the key risk change moment; S52, trigger three-dimensional early warning display: when the risk change amount exceeds 0.2 or reaches the key risk change moment, automatically trigger the three-dimensional early warning display, and real-time display the patient's risk state through three-dimensional visualization graphics; S53, generate personalized intervention scheme: according to the results of three-dimensional early warning display, combined with the personalized information of the patient, generate a personalized intervention scheme, including drug treatment, rehabilitation training, and diet adjustment.

Images