Search

CN-122024831-A - Sudden cardiac death early warning method based on wearable device signals and gene markers

CN122024831ACN 122024831 ACN122024831 ACN 122024831ACN-122024831-A

Abstract

The invention provides a sudden cardiac death early warning method based on wearable equipment signals and gene markers, which comprises the steps of extracting user history medical records and full genome SNP loci by natural language processing and bioinformatics algorithm, screening to obtain key gene markers, adopting wearable equipment multidimensional physiological signals and clustering state intervals, establishing a gene-physiological heterogeneous coupling map by combining gene and physiological parameters through cross-modal mutual information and causal analysis, carrying out side weight update and sparse screening by using a dynamic map learning and minimum spanning tree method with L1 regularization, extracting a core interaction path, constructing a dynamic risk assessment equation based on symbol regression and Bayesian optimization technology, carrying out individual risk classification and interactive chain attribution on real-time physiological data, and finally generating a multi-level visual report for providing visual risk judgment and decision support for clinicians.

Inventors

  • LI SI
  • LIU CHENGWU

Assignees

  • 中山大学附属第一医院
  • 广州卫生职业技术学院

Dates

Publication Date
20260512
Application Date
20260204

Claims (10)

  1. 1. The sudden cardiac death early warning method based on the wearable device signal and the gene marker is characterized by comprising the following steps of: S1, extracting SNP locus information related to sudden cardiac death based on a user history electronic medical record and whole genome sequencing data, screening to form a gene marker set, and synchronously dividing physiological characteristic intervals of multidimensional physiological signals acquired by wearable equipment; s2, calculating cross-modal mutual information and Grangel causal strength according to the gene marker set and the divided physiological state interval, and constructing an initial heterogeneous graph structure containing gene-physiological node interaction; s3, based on a dynamic graph learning mechanism of L1 regularization constraint, performing edge weight updating and weak connection cutting on the initial iso-graph structure by adopting a minimum spanning tree strategy to generate a sparse gene-physiological coupling graph; s4, aiming at a core sub-network in the sparse gene-physiological coupling map, searching a nonlinear mathematical expression by applying a symbolic regression technology to generate an interpretable risk evolution equation; s5, inputting the physiological signal characteristic values acquired in real time into the interpretable risk evolution equation, calculating a dynamic risk score increment, generating a current risk level by combining a baseline risk value, and identifying a dominant gene-physiological interaction chain through a decision tree model; S6, generating a multi-level visual report based on the current risk level and the contribution trend of the gene-physiological interactive chain.
  2. 2. The sudden cardiac death warning method based on a wearable device signal and a genetic marker according to claim 1, wherein the step S1 specifically comprises: Based on the user history electronic medical record data, executing a natural language processing and structured information extraction algorithm, extracting history diagnosis records, medication information and clinical events related to sudden cardiac death, and generating a structured medical history feature vector; According to the whole genome sequencing data, performing quality control and annotation analysis on an original VCF file by adopting a bioinformatics tool, identifying SNP loci known to be obviously related to sudden cardiac death, and generating a candidate SNP set; Based on the candidate SNP set, performing gene pathway enrichment analysis and multivariate logistic regression model screening, identifying gene markers with obvious functional influence in the electrophysiological activity of the heart, and constructing an individualized gene marker set; collecting multi-dimensional physiological signals by using wearable equipment, and executing time-frequency domain feature extraction and multi-mode fusion processing on the multi-dimensional physiological signals to generate high-vitamin feature vectors; based on the high-vitamin physiological feature vector, a clustering algorithm is adopted to perform state division on the continuous time sequence data, physiological feature intervals of the user in resting, movement and sleep states are identified and marked, and a state label sequence is generated.
  3. 3. The method of claim 2, wherein the multi-dimensional physiological signals include electrocardiogram, heart rate variability, accelerometer and respiratory rate signals.
  4. 4. The sudden cardiac death warning method based on a wearable device signal and a genetic marker according to claim 1, wherein the step S2 specifically comprises: performing a gene-phenotype association analysis based on the SNP site information in the user gene marker set, identifying a node of genetic variation significantly associated with sudden cardiac death; performing time-frequency domain feature extraction processing on the multidimensional physiological signals acquired by the wearable equipment to obtain key physiological feature parameters; based on the gene marker node and the extracted physiological characteristic parameters, performing cross-mode mutual information calculation processing; Based on the time sequence data of the key physiological characteristic parameters, executing a Granges causal analysis algorithm, identifying characteristic pairs with causal relation in different physiological states, and constructing a causal directivity matrix; Based on the cross-modal mutual information result and the causal directivity matrix, executing heterogeneous graph structure modeling processing to generate an initial gene-physiological coupling map composed of gene nodes and physiological characteristic nodes.
  5. 5. The method for early warning sudden cardiac death based on wearable device signals and gene markers according to claim 4, wherein the step S2 further comprises measuring statistics and time sequence causal dependencies between genes and physiological characteristic nodes respectively by using cross-modal mutual information calculation and grange causal analysis, generating an initial heterogram adjacency matrix by using joint linear weighting and regularization, fusing node attributes and side weight data, and supporting multi-node high-dimensional interaction and significance screening of side weights by using a graph structure.
  6. 6. The sudden cardiac death warning method based on a wearable device signal and a genetic marker according to claim 1, wherein the step S3 specifically comprises: Based on the initial heterogeneous graph structure and physiological signal data in a latest 24-hour window, calculating a dynamic mutual information increment between a gene marker node and a physiological characteristic node to obtain a dynamic mutual information matrix; Based on the dynamic mutual information matrix and the initial value of the grange causal strength, performing weighted fusion operation, and dynamically updating the edge weights among nodes by adopting a sliding window normalization method to construct an updated gene-physiological abnormal composition adjacency matrix; based on the updated gene-physiological abnormal graph adjacent matrix, applying a graph learning algorithm of L1 regularization constraint to execute sparsification processing on all sides in the graph to obtain a sparse graph structure candidate set; Based on the sparse graph structure candidate set, executing a minimum spanning tree algorithm, and extracting a trunk sub-graph containing a key gene-physiological interaction path; based on the trunk subgraph, topology structure optimization and weight recalibration operation are executed, and local readjustment is carried out on the edge weight by adopting a weighting strategy based on node degree and path length, so that a final sparse gene-physiological coupling map is generated.
  7. 7. The sudden cardiac death warning method based on a wearable device signal and a genetic marker according to claim 1, wherein the step S4 specifically comprises: based on a core subnetwork in a sparse gene-physiological coupling map, extracting gene nodes closely related to sudden cardiac death and physiological characteristic nodes obviously coupled with the gene nodes, and taking the gene nodes as an input variable set of symbolic regression modeling to construct a basic variable space of an interpretable risk evolution model; Carrying out characteristic engineering treatment on SNP load, HRV reduction rate and interaction items thereof in the input variable set to generate a basic function library; executing symbolic regression based on a genetic programming algorithm, searching an optimal nonlinear expression structure from the basic function library to minimize the sum of squares of residual errors between model output and historical risk scores, and obtaining a preliminary risk evolution equation prototype; constructing a Bayesian optimization framework, setting parameter prior distribution constraint conditions based on an action mechanism of SNP load and HRV reduction rate in the electrophysiological stability of the heart in a prior biological knowledge base, and performing posterior estimation optimization on parameters in the risk evolution equation prototype; and executing model verification and sparsification processing on the risk evolution equation after Bayesian optimization to generate a final interpretable dynamic risk modeling equation.
  8. 8. The sudden cardiac death warning method based on a wearable device signal and a genetic marker according to claim 1, wherein the step S5 specifically comprises: Based on multidimensional physiological signals acquired by wearable equipment in real time, extracting an electrocardiogram R-R interval sequence and heart rate variability time-frequency domain characteristic parameters to obtain an HRV reduction rate index; carrying out standardization processing on SNP load values extracted from sequencing data of the whole genome of an individual, and calculating an individualized static gene risk baseline value by utilizing a gene risk coefficient constrained by a Bayesian optimization framework; inputting the HRV reduction rate and the SNP load value into a nonlinear risk evolution equation generated by symbolic regression, and executing dynamic product and exponential operation based on equation parameters to obtain a real-time risk increment value; performing weighted fusion processing on the static gene risk baseline value and the real-time risk increment value to generate an individualized current dynamic risk score; A decision tree model is constructed based on the sparse gene-physiological coupling map, the current risk score and the activation intensity of each gene-physiological node are input, the dominant gene-physiological interaction chain is identified through path attribution analysis, and the contribution weight and the action direction of the dominant gene-physiological interaction chain in the current risk evolution are output.
  9. 9. The sudden cardiac death warning method based on wearable device signals and gene markers according to claim 8, wherein the step S5 further comprises weighting and fusing historical risk baselines and risk increments based on a physiological state adaptive weighting algorithm by utilizing physiological signal characteristics acquired in real time and gene risk baselines input interpretable equations, modeling current risk scores and node activation intensities in combination with CART decision trees, and outputting dominant gene-physiological interaction chains and weights, action directions and quantitative contributions to dynamic scores thereof through splitting gain and mutual information increasing and decreasing sign path analysis.
  10. 10. The sudden cardiac death warning method based on the wearable device signal and the gene marker according to claim 1, wherein in the step S6, the current risk level adopts five-color gradient coding, the risk level is coded into five-color gradient output of red, orange, yellow, blue and green, and the interactive chain path is dynamically rendered through a topological map and annotated with biological path annotation information.

Description

Sudden cardiac death early warning method based on wearable device signals and gene markers Technical Field The invention relates to the technical field of dynamic assessment of sudden cardiac death risk, in particular to a sudden cardiac death early warning method based on wearable equipment signals and gene markers. Background Along with gradual fusion of dynamic physiological signals of wearable equipment and individual genome information of a sudden cardiac death risk prediction and early warning system, various risk modeling technologies based on data driving and deep learning models emerge in the field of medical artificial intelligence and bioinformatics intersection. The existing mainstream scheme generally adopts a large-scale neural network model to carry out joint modeling on the dynamically acquired electrocardio and physiological signals and known gene markers, and the dynamic prediction on the sudden cardiac death risk is realized by learning a complex nonlinear mode to distribute weights. In recent years, part of technologies further integrate an attention mechanism and an adaptive weight updating algorithm, and the representation capability and prediction accuracy of the model on multi-mode data are improved. In addition, in the modeling process, partial products also try to embed genome SNP loci and physiological characteristics in a combined way in a multi-layer network structure, so that individuation risk quantification is realized; such artificial intelligence risk prediction systems have been widely used in medical pre-warning, intelligent auxiliary diagnosis, and individual health management scenarios. The development trend is mainly characterized in that the breadth of data fusion is increased, the complexity of a model is improved, and the real-time risk assessment based on dynamic signals is gradually realized. Typical technical applications include risk discrimination of electrocardiographic state sequences based on deep learning (e.g., LSTM, GRU), genetic risk stratification by combining variant site heatmaps, and optimizing prediction of sudden death event occurrence probability by multi-modal feature fusion; However, the main stream technical scheme has obvious defects that firstly, although the dynamic risk model based on deep learning improves the retrospective prediction accuracy of cases, the decision process of the dynamic risk model often belongs to the category of black boxes and lacks a transparent causal interpretation mechanism. The weight distribution result is difficult to trace back to a specific gene-physiological mechanism path, and medical staff cannot know the biological reasons for the increase or decrease of the model judgment risk. The method greatly reduces the clinical credibility and the adoption willingness of the method and affects the actual medical application, and secondly, the existing dynamic weight adjustment method mostly depends on global attention or naive fusion, and the causal sparsity and the path coupling characteristics in the real biological network are ignored. The dynamic effect of physiological state changes on gene risk expression is difficult to obtain clear characterization in a model structure, and more importantly, the lack of a mechanism for explaining a dominant interaction path is unfavorable for disease mechanism research and establishment of a personalized intervention strategy. Disclosure of Invention The invention aims to solve the technical problems and provides a sudden cardiac death early warning method based on a wearable device signal and a gene marker. The technical scheme of the invention is realized by a sudden cardiac death early warning method based on a wearable device signal and a gene marker, which comprises the following steps: s1, extracting SNP locus information related to sudden cardiac death based on a user history electronic medical record and whole genome sequencing data, screening to form a gene marker set, and synchronously dividing multidimensional physiological signals acquired by wearable equipment into rest/movement/sleep state intervals; S2, calculating cross-modal mutual information and Grandide causal strength according to the gene marker set and the divided physiological state interval, and constructing an initial heterogram structure containing gene-physiological node interaction, wherein the gene marker nodes are connected with physiological feature nodes through dynamic weight edges; S3, based on a dynamic graph learning mechanism of L1 regularization constraint, performing edge weight updating and weak connection cutting on an initial heterogeneous graph structure by adopting a minimum spanning tree strategy, generating a sparse gene-physiological coupling map every 24 hours period, and reserving a dynamic key interaction path; S4, searching a nonlinear mathematical expression by applying a symbolic regression technology aiming at a core subnetwork in the sparse gene-physiological coupling