Medical risk factor causal relationship extraction method based on double-constraint guidance

Abstract

The invention belongs to the technical field of medical artificial intelligence, and provides a medical risk factor causal relationship extraction method based on double-constraint guidance. The method comprises the following steps: firstly, preprocessing clinical medical data containing patient disease categories and physiological index risk factors, and then constructing an MI constraint matrix for quantifying variable dependence and an M constraint matrix for describing an association relationship by using mutual information and an MGM method respectively; initializing a multi-linear / non-linear causal structure learning device by using the M constraint matrix, and obtaining a plurality of causal intensity matrixes in combination with double constraint correction; and generating a consensus causal intensity matrix through a aggregation operator, and finally obtaining a clear causal structure between the disease and the risk factor after ring removal processing and threshold interception. The method is high in precision and generalization ability, can effectively identify the core causal relationship, and provides reliable technical support for clinical cause inference and accurate intervention.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence technology, and in particular relates to a method for extracting causal relationships of medical risk factors based on dual-constraint guidance. Background Technology

[0002] In the wave of intelligent transformation in the healthcare field, accurately identifying the causal relationship between diseases and risk factors is one of the core directions of precision medicine research. The accumulation of massive amounts of electronic health records, omics data, and clinical data provides a data foundation for exploring the intrinsic mechanisms of disease occurrence and development. Clear causal relationships can provide a scientific basis for clinical etiological inference, precise intervention design, and public health decision-making, directly affecting the effectiveness and targeting of medical services, and having significant practical implications for improving the overall health level of the population.

[0003] Current disease risk factor identification technologies are mainly divided into two categories: one is based on correlation analysis models, such as logistic regression, random forests, and deep neural networks, which screen potential risk factors by ranking feature importance. However, these methods can only reveal statistical associations between variables and cannot determine the causal direction. They are difficult to distinguish between key issues such as "risk factors cause disease" and "disease causes changes in related indicators," and have fundamental defects in cross-group generalization, intervention effect prediction, and mechanism explanation. The other category is causal structure learning techniques, including constraint-based methods (such as PC and FCI algorithms) and scoring-based methods (such as GES algorithms). Although these aim to infer causal structures, they face many challenges when processing real-world medical data. On high-dimensional, sparse, nonlinear datasets with a large amount of potential confounding, they are prone to problems such as search space explosion, inaccurate identification of causal direction, and sensitivity to noise, resulting in unstable causal network structures, poor interpretability, and significant deficiencies in prior knowledge fusion and causal strength quantification.

[0004] The core shortcomings of existing technologies limit their practical application in clinical scenarios, specifically manifested in two key issues: First, the lack of an effective guidance mechanism makes it difficult for existing algorithms to effectively integrate the rich biological pathways, literature evidence, and expert experience in the medical field, leading to a blind and inefficient causal search process that is prone to getting trapped in local optima. Second, insufficient optimization of causal effect weights results in a lack of further guidance on potential causal relationships during iterative optimization, and a lack of refined assessment and weighted optimization mechanisms for the causal strength of different risk factors, making it difficult to accurately distinguish between core driving factors and secondary associated variables. Therefore, there is an urgent need for a novel causal identification framework that integrates domain-constrained guidance and data-driven learning to overcome the bottlenecks of existing technologies and improve the accuracy and reliability of identifying causal relationships of disease risk factors. Summary of the Invention

[0005] To address the problems existing in the background art, one aspect of the present invention proposes a method for extracting causal relationships of medical risk factors based on dual-constraint guidance, comprising:

[0006] S1: Obtain the preprocessed clinical medical dataset, which includes: the patient's disease category and the patient's various physiological indicators, wherein the patient's various physiological indicators are regarded as risk factors;

[0007] S2: Based on clinical medical datasets, mutual information is used to quantify the relationship between diseases and risk factors, as well as the dependencies between risk factors, to construct an MI constraint matrix;

[0008] S3: Based on clinical medical datasets, the MGM method is used to quantify the relationship between diseases and risk factors, as well as the association between risk factors, and to construct an M-constraint matrix;

[0009] S4: The M-constraint matrix is ​​used as the initial weight matrix and input into multiple causal structure learners for derivation. The consistency and heterogeneity of the M-constraint matrix and the MI-constraint matrix are used to correct the causal structure to obtain multiple causal strength matrices.

[0010] S5: Aggregate multiple causal strength matrices using aggregation operators to obtain a consensus causal strength matrix;

[0011] S6: The final causal structure between disease risk factors is obtained by removing loops from the consensus causal strength matrix and thresholding.

[0012] Another aspect of the present invention provides a medical risk factor causal relationship extraction system based on dual constraint guidance, the system comprising a memory and a processor; the memory is used to store an application program; the processor is used to run the application program and execute the aforementioned medical risk factor causal relationship extraction method based on dual constraint guidance.

[0013] Another aspect of the present invention provides a computer storage medium storing a remote monitoring program, which, when executed by a processor, implements the aforementioned method for extracting causal relationships of medical risk factors based on dual-constraint guidance.

[0014] The present invention has at least the following beneficial effects

[0015] This invention employs an innovative design that combines dual-constraint guidance with integrated causal structure learning. This approach effectively incorporates potential dependency information from medical data, narrowing the causal search space and avoiding blind iteration. Furthermore, it enhances adaptability and generalization capabilities for high-dimensional, sparse, and nonlinear medical data by integrating multilinear / nonlinear causal structure learners and generating a consensus causal strength matrix through aggregation operators. Finally, loop removal ensures the acyclic logic of the causal graph, and threshold truncation filters strong causal relationships. Ultimately, this achieves a precise leap from "statistical association" to "explicit causality," enabling more accurate identification of the core causal relationships between diseases and risk factors. This provides scientific and reliable technical support for clinical etiology inference, precision intervention design, and public health decision-making, overcoming the shortcomings of existing methods in interpretability, stability, and refined assessment of causal strength. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0017] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0018] Please see Figure 1 One aspect of this invention proposes a method for extracting causal relationships of medical risk factors based on dual-constraint guidance, comprising:

[0019] S1: Obtain the preprocessed clinical medical dataset, which includes: the patient's disease category and the patient's various physiological indicators, wherein the patient's various physiological indicators are regarded as risk factors;

[0020] Preferably, the preprocessing includes:

[0021] S11: Clinical medical datasets can be represented as... ,in, , ,in, express The first in One risk factor, This represents a sample of clinical medical data; express Disease variables; Indicates the quantity of risk factors;

[0022] S12: If clinical medical data samples Missing number Risk factors Then calculate the clinical medical data sample. Compared with other non-missing data in the clinical medicine dataset The similarity of clinical medical data samples for each risk factor was used to select the Top-K most similar samples in the second step. The mean or median of each risk factor in a sample of clinical medical data. Risk factors Fill in the gaps;

[0023] S13: Calculate the number of clinical medical data samples in the clinical medical dataset at the [number]th [time]. Mean of each risk factor and standard deviation If a certain clinical medical data sample The Risk factors k represents a user-defined positive integer; then the clinical medical data sample The Risk factors Replace with mean .

[0024] In this embodiment, the clinical medical dataset can be obtained through hospital information systems, laboratory information systems, electronic medical record systems, clinical data centers, etc., or through public platforms such as the National Biobank Public Service Platform, the Chinese Academy of Medical Sciences Medical Data Sharing Platform, and the Regional Medical and Health Data Open Platform. Other methods are also possible. The acquired data includes various types of diseases and multiple physiological indicators of patients. These physiological indicators mainly include multiple indicators such as age, gender, body temperature, blood pressure, heart rate, respiratory rate, complete blood count, weight and BMI, electrocardiogram, pulmonary function indicators, and bone density. The specific indicators can be selected by those skilled in the art. It is important to understand that the clinical medical dataset contains a large amount of disease-related data. If causal analysis is needed for a target disease, the disease variables in the clinical medical data sample must be modified. For example, for leukemia, if a patient has leukemia, the disease variables in the clinical medical data sample must be changed. The disease variable in the clinical medical data sample is changed to "Yes" if the patient has other diseases but not leukemia. Conversely, if the patient has other diseases but not leukemia, the disease variable in the clinical medical data sample is changed to "No". Similarly, the same processing can be performed on the clinical medical dataset for other target diseases. Examples are not provided in this article. Disease variables do not necessarily have to be defined as "Yes" or "No"; they can also be defined as 0 or 1. Furthermore, for a physiological indicator, different patients may have the same or different values. Therefore, when calculating mutual information, it is necessary to sum and statistically analyze different physiological values ​​under the same physiological indicator. In this embodiment, to address the common problem of missing values ​​in medical data, the missing values ​​are filled by finding multiple samples most similar to the missing data sample and using the mean or median of the corresponding indicators of these similar samples. This preserves the original intrinsic correlation characteristics of the data to the greatest extent, which is more in line with the real data patterns than simply filling in fixed values. Finally, for outliers in the data (such as extreme physiological indicators that far exceed most samples), replacing these outliers with the overall mean of the indicator can effectively eliminate the interference of extreme noise on the data and avoid outliers distorting the true dependencies between variables.

[0025] S2: Based on clinical medical datasets, mutual information is used to quantify the relationship between diseases and risk factors, as well as the dependencies between risk factors, to construct an MI constraint matrix;

[0026] Preferably, constructing the MI constraint matrix includes:

[0027]

[0028]

[0029] in, Represents the first in the MI constraint matrix Line 1 Column elements; and This indicates a disease and its risk factors, or different risk factors. This indicates the set threshold. Indicating in clinical medical datasets and The probability of them occurring simultaneously; Indicating in clinical medical datasets The probability of occurrence; Indicating in clinical medical datasets The probability of occurrence. In this embodiment, step S2 quantifies the dependency relationship between the disease and risk factors, and between risk factors, and constructs the MI constraint matrix through mutual information. This provides key data-driven constraint support for subsequent causal relationship extraction, with significant beneficial effects: On the one hand, mutual information can accurately measure the "degree of information sharing" between variables. Regardless of whether the relationship between variables is linear or nonlinear, it can effectively capture the strength of their dependency, perfectly adapting to the complex variable relationships in medical data and avoiding the omission of nonlinear relationships by traditional linear analysis; on the other hand, by setting a threshold to filter out strongly dependent variable pairs, it can effectively filter out noise interference and meaningless weak associations in the data, narrowing the search range for subsequent causal structure learning in advance, avoiding blind iteration of the algorithm in a massive number of irrelevant variable pairs, and improving the efficiency of analysis.

[0030] S3: Based on clinical medical datasets, the MGM method is used to quantify the relationship between diseases and risk factors, as well as the association between risk factors, and to construct an M-constraint matrix;

[0031] In this embodiment, MGM (Hybrid Graph Model) is a technique used to characterize the relationships between multiple variables. Its core logic is to analyze the potential relationships between variables through "local conditional independence testing": when analyzing the relationship between two target variables (such as a risk factor and a disease, or two different risk factors), the influence of all other variables is controlled. If, after control, the two target variables still do not satisfy the independence condition (i.e., their relationship is not affected by other variables), then a stable potential relationship exists between them, and these two variables are connected in the graph with "undirected edges." If, after controlling other variables, the two become independent, then there is no relationship, and no edges are drawn in the graph. Ultimately, a graph structure of "variable-undirected edges" (i.e., a causal skeleton) is constructed in this way, and then this structure is transformed into an M-constraint matrix—in the matrix, "1" represents that the corresponding variable pair has the aforementioned stable relationship (potential causal possibility), and "0" represents no relationship (excluding causal possibility).

[0032] Step S3 constructs an M-constraint matrix using the MGM method, providing crucial "association skeleton guidance" for subsequent causal relationship extraction. First, MGM overcomes the limitations of isolated pairwise variable analysis, determining associations while controlling for interference from other variables. This better aligns with the complex scenarios in medical data where "multiple risk factors interact and jointly affect the disease" (e.g., when analyzing the association between blood glucose and diabetes, interference from variables like weight and blood lipids can be excluded, ensuring more accurate association judgment). Second, the constructed M-constraint matrix clearly defines "potentially causal variable pairs" and "non-causal variable pairs" using "1 / 0," effectively defining an "effective search range" for subsequent causal structure learning. This directly filters out unrelated variable pairs, preventing the algorithm from blindly iterating through massive combinations of irrelevant variables and significantly improving subsequent learning efficiency. Finally, the "causal skeleton" formed by this matrix serves as initial guidance for subsequent causal structure learning, transforming the learning process from "undirected exploration" to "precise optimization guided by a skeleton." This lays the foundation for subsequent "dual-constraint correction" using the MI constraint matrix, further ensuring the accuracy of causal relationship identification.

[0033] S4: The M-constraint matrix is ​​used as the initial weight matrix and input into multiple causal structure learners for derivation. The consistency and heterogeneity of the M-constraint matrix and the MI-constraint matrix are used to correct the causal structure to obtain multiple causal strength matrices.

[0034] Preferably, the plurality of causal structure learners include: one or more linear causal structure learners of different types, and one or more nonlinear causal structure learners of different types.

[0035] In this embodiment, the linear causal structure learners mainly include existing causal structure learners such as the PC algorithm, GES algorithm, DirectLiNGAM / ICALiNGAM (linear non-Gaussian) algorithm, and NOTEARS (gradient-based linear algorithm). These are existing linear causal structure learners, and their derivation process and working principle should be understood by those skilled in the art. Therefore, they will not be described in more detail in this paper. The non-linear causal structure learners mainly include existing causal structure learners such as the FCI algorithm, Kernel-PC algorithm, ANM algorithm, NOTEARS-MLP / GraNDAG algorithm, and ICD algorithm. These are existing linear causal structure learners, and their derivation process and working principle should be understood by those skilled in the art. Therefore, they will not be described in more detail in this paper.

[0036] Preferably, the modification of the causal structure using the consistency and heterogeneity of the M-constraint matrix and the MI-constraint matrix includes:

[0037] S41: Use the M-constraint matrix as the initial weight matrix and input it into multiple causal structure learners for derivation;

[0038] S42: During the continuous optimization process of each causal structure learner, the derivation results obtained in each iteration are corrected using the consistency and heterogeneity of the M constraint matrix and the MI constraint matrix, until the causal structure learner converges or reaches the preset number of iterations. The correction using the consistency and heterogeneity of the M constraint matrix and the MI constraint matrix includes:

[0039]

[0040] in, Indicates the first The causal structure learner in the first... The causal strength of a and b after correction in the next iteration; No. The causal structure learner in the first... The causal strength of a and b in the next iteration; This indicates the set weighting factor; This represents the element in the a-th row and b-th column of the M-constraint matrix; This represents the element in the a-th row and b-th column of the MI constraint matrix; elements a and b represent the disease and risk factors, or different risk factors.

[0041] In this embodiment, step S4 employs a core strategy of "using the M-constraint matrix as initial guidance, parallel derivation of multiple types of learners, and dynamic correction of dual constraints" to provide accurate and comprehensive intermediate results support for causal relationship extraction. First, the M-constraint matrix is ​​used as the initial weight matrix input to the learner, rather than being randomly initialized. This allows the learning process to start directly from a clearly defined "causal framework," avoiding the aimless blind search of traditional algorithms, significantly reducing ineffective iterations, and substantially improving the efficiency of causal learning. Second, the parallel derivation of multiple causal structure learners, including linear and nonlinear types, can simultaneously adapt to medical data... Complex causal relationship patterns—covering both "direct linear correlation between a certain physiological indicator and a disease" and "nonlinear correlation of multiple indicators working together to cause a disease"—avoid omissions of specific types of relationships by a single learner, ensuring the comprehensiveness of the results. Finally, the results of each iteration are dynamically corrected through dual constraints (M constraint and MI constraint): correlations supported by both constraints (strong potential causal signals) are weighted and enhanced, while correlations denied by both constraints (no causal possibility) are directly set to 0. This strengthens reliable causal signals and filters out meaningless noisy correlations, ensuring that the causal strength matrix output by each learner is more accurate.

[0042] S5: Aggregate multiple causal strength matrices using aggregation operators to obtain a consensus causal strength matrix;

[0043] Preferably, the aggregation of multiple causal intensity matrices using the aggregation operator includes:

[0044] S51: Calculate the distance between any two causal strength matrices using the Euclidean distance formula, and calculate the support between the two causal strength matrices based on the distance between them.

[0045] Preferably, the support between the causal strength matrices includes:

[0046]

[0047] in, Represents the causal strength matrix Relative to the causal strength matrix Support level; Represents the causal strength matrix and causal strength matrix Euclidean distance;

[0048] S52: Obtain the total support of each causal strength matrix based on the support between the two causal strength matrices;

[0049] Preferably, the total support of each causal strength matrix includes:

[0050]

[0051] Where T represents the number of causal strength matrices; Represents the causal strength matrix Total support;

[0052] S53: Aggregate all causal strength matrices using the MPA aggregation operator based on the total support of each causal strength matrix to obtain the consensus causal strength matrix;

[0053] Preferably, the consensus causal strength moment includes:

[0054]

[0055]

[0056] in, Represents the normalization coefficient; Represents the causal strength matrix The weights; Indicates the regulating factor; Represents the causal strength matrix The element in row a and column b of the array; Represents the consensus causal strength moment The element in row a and column b of the array.

[0057] In this embodiment, step S5 aggregates multiple causal strength matrices using an aggregation operator to obtain a consensus causal strength matrix. This effectively solves the potential bias problem of a single causal structure learner, providing a more stable and reliable intermediate result for subsequent causal relationship confirmation. First, it breaks the limitation of relying on the output of a single learner—different learners (linear or nonlinear) have different adaptability to complex medical data and may have their own analytical biases. Aggregation can combine the advantages of multiple learners, avoiding the one-sidedness of a single result and making the characterization of causal strength more comprehensive. Second, the aggregation process first uses Euclidean algorithm... Distance measures the similarity between different matrices and calculates support, which can accurately determine the credibility of each causal strength matrix. This allows matrices with high support to have higher weights during aggregation, significantly reducing the interference of abnormal or unreliable matrices on the results. Finally, weighted fusion is achieved using the MPA aggregation operator, rather than simple averaging, which further strengthens the contribution of the credible matrix. The resulting consensus causal strength matrix integrates the effective information from multiple learners and filters out noise interference. Its stability and accuracy far exceed those of a single matrix, providing a high-quality foundation for subsequent loop removal and thresholding, and directly improving the robustness of the final causal relationship identification.

[0058] S6: The final causal structure between disease risk factors is obtained by removing loops from the consensus causal strength matrix and thresholding.

[0059] Preferably, step S6 includes:

[0060] S61: Construct a directed graph from cause to effect by taking the rows of the consensus causality strength matrix as causes and the columns as results;

[0061] S62: Use the DFS method to traverse the directed graph G. If a node that is already in the path is encountered during the traversal, it is determined that there is a cycle. Remove the edge with the smallest weight in the cycle until DFS can no longer detect any cycles and obtains an acyclic consensus matrix.

[0062] S63: Set a threshold β, and use the threshold cutoff of the acyclic consensus matrix to obtain the final causal structure between disease risk factors.

[0063] In this embodiment, step S6 is further explained: consensus causality strength matrix. , 1 represents the number of risk factor variables, and 1 represents the number of disease variables; Representing variables For variables The causal strength is defined by the directed graph. , where the node set Each node corresponds to a variable, and there is a directed edge set E. Then at node With nodes Add an edge between them, with the edge weight equal to Traverse the directed graph using Depth-First Search (DFS). The process records the access path starting from the initial node. If a node already in the path is encountered during traversal, a cycle is determined to exist. The edge with the smallest weight in the cycle is removed, and this process continues until DFS detects no cycles and obtains a cycle-free consensus matrix. Set a threshold β and define the final causal matrix. The rules for element value selection are as follows:

[0064]

[0065] in, Represents the final causal matrix Medium variables For variables The ultimate causal relationship, that is, the ultimate causal structure among disease risk factors.

[0066] In this embodiment, through a series of operations—"constructing a directed graph, performing DFS loop removal, and thresholding"—the abstract consensus causal strength matrix is ​​transformed into a final causal structure that conforms to causal logic and has practical application value. First, a directed graph is constructed by "rows corresponding to causes and columns corresponding to results," transforming the numerical causal strength in the matrix into an intuitive "node-directed edge" relationship. Second, DFS loop removal accurately identifies and removes the weakest edge in the causal loop, completely eliminating the logical contradiction of "cause → result → cause," ensuring that the final causal structure conforms to the causal essence of "directed loop-free," and avoiding erroneous causal logic from misleading clinical judgment. Finally, thresholding removes weak associations with extremely low causal strength, retaining only strong causal relationships, effectively filtering out meaningless noise associations, making the output "disease-risk factor" causal relationship more focused on the core driving factors, and directly providing clear and reliable evidence for clinical etiology analysis and precise intervention design, significantly improving the practicality and guidance of the causal relationship extraction results.

[0067] Another aspect of the present invention provides a medical risk factor causal relationship extraction system based on dual constraint guidance, the system comprising a memory and a processor; the memory is used to store an application program; the processor is used to run the application program and execute the aforementioned medical risk factor causal relationship extraction method based on dual constraint guidance.

[0068] Another aspect of the present invention provides a computer storage medium storing a remote monitoring program, which, when executed by a processor, implements the aforementioned method for extracting causal relationships of medical risk factors based on dual-constraint guidance.

[0069] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0070] In summary, this invention, through an innovative design combining dual-constraint guidance and integrated causal structure learning, effectively incorporates potential dependency information from medical data, narrowing the causal search space and avoiding blind iteration. Furthermore, it enhances adaptability and generalization capabilities for high-dimensional, sparse, and nonlinear medical data by integrating multilinear / nonlinear causal structure learners and generating a consensus causal strength matrix using aggregation operators. Finally, loop removal ensures the acyclic logic of the causal graph, and threshold truncation filters strong causal relationships, ultimately achieving a precise leap from "statistical association" to "explicit causality." This enables more accurate identification of the core causal relationships between diseases and risk factors, providing scientific and reliable technical support for clinical etiology inference, precision intervention design, and public health decision-making. It also addresses the shortcomings of existing methods in interpretability, stability, and refined evaluation of causal strength.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting causal relationships of medical risk factors based on dual-constraint guidance, characterized in that, include: S1: Obtain the preprocessed clinical medical dataset, which includes: the patient's disease category and the patient's various physiological indicators, wherein the patient's various physiological indicators are regarded as risk factors; S2: Based on clinical medical datasets, mutual information is used to quantify the relationship between diseases and risk factors, as well as the dependencies between risk factors, to construct an MI constraint matrix; S3: Based on clinical medical datasets, the MGM method is used to quantify the relationship between diseases and risk factors, as well as the association between risk factors, and to construct an M-constraint matrix; S4: The M-constraint matrix is ​​used as the initial weight matrix and input into multiple causal structure learners for derivation. The consistency and heterogeneity of the M-constraint matrix and the MI-constraint matrix are used to correct the causal structure to obtain multiple causal strength matrices. S5: Aggregate multiple causal strength matrices using aggregation operators to obtain a consensus causal strength matrix; S6: The final causal structure between disease risk factors is obtained by removing loops from the consensus causal strength matrix and thresholding.

2. The method for extracting causal relationships of medical risk factors based on dual-constraint guidance according to claim 1, characterized in that, The preprocessing includes: S11: Clinical medical datasets can be represented as... ,in, , ,in, express The first in One risk factor, This represents a sample of clinical medical data; express Disease variables; Indicates the quantity of risk factors; S12: If clinical medical data samples Missing number Risk factors Then calculate the clinical medical data sample. Compared with other non-missing data in the clinical medicine dataset The similarity of clinical medical data samples for each risk factor was used to select the Top-K most similar samples in the second step. The mean or median of each risk factor in a sample of clinical medical data. Risk factors Fill in the gaps; S13: Calculate the number of clinical medical data samples in the clinical medical dataset at the [number]th [time]. Mean of each risk factor and standard deviation If a certain clinical medical data sample The Risk factors k represents a user-defined positive integer; then the clinical medical data sample The Risk factors Replace with mean .

3. The method for extracting causal relationships of medical risk factors based on dual-constraint guidance according to claim 1, characterized in that, The construction of the MI constraint matrix includes: in, Represents the first in the MI constraint matrix Line number Column elements; and This indicates a disease and its risk factors, or different risk factors. This indicates the set threshold. Indicating in clinical medical datasets and The probability of them occurring simultaneously; Indicating in clinical medical datasets The probability of occurrence; Indicating in clinical medical datasets The probability of occurrence.

4. The method for extracting causal relationships of medical risk factors based on dual-constraint guidance according to claim 1, characterized in that, The plurality of causal structure learners include: one or more linear causal structure learners of different types, and one or more nonlinear causal structure learners of different types.

5. The method for extracting causal relationships of medical risk factors based on dual-constraint guidance according to claim 1, characterized in that, The method of modifying the causal structure using the consistency and heterogeneity of the M-constraint matrix and the MI-constraint matrix includes: S41: Use the M-constraint matrix as the initial weight matrix and input it into multiple causal structure learners for derivation; S42: During the continuous optimization process of each causal structure learner, the derivation results obtained in each iteration are corrected using the consistency and heterogeneity of the M constraint matrix and the MI constraint matrix, until the causal structure learner converges or reaches the preset number of iterations. The correction using the consistency and heterogeneity of the M constraint matrix and the MI constraint matrix includes: in, Indicates the first The causal structure learner in the first... The causal strength of a and b after correction in the next iteration; No. The causal structure learner in the first... The causal strength of a and b in the next iteration; This indicates the set weighting factor; This represents the element in the a-th row and b-th column of the M-constraint matrix; This represents the element in the a-th row and b-th column of the MI constraint matrix; elements a and b represent the disease and risk factors, or different risk factors.

6. The method for extracting causal relationships of medical risk factors based on dual-constraint guidance according to claim 1, characterized in that, The aggregation of multiple causal intensity matrices using aggregation operators includes: S51: Calculate the distance between any two causal strength matrices using the Euclidean distance formula, and calculate the support between the two causal strength matrices based on the distance between them. S52: Obtain the total support of each causal strength matrix based on the support between the two causal strength matrices; S53: Aggregate all causal strength matrices using the MPA aggregation operator based on the total support of each causal strength matrix to obtain the consensus causal strength matrix.

7. The method for extracting causal relationships of medical risk factors based on dual-constraint guidance according to claim 1, characterized in that, Step S6 includes: S61: Construct a directed graph from cause to effect by taking the rows of the consensus causality strength matrix as causes and the columns as results; S62: Use the DFS method to traverse the directed graph G. If a node that is already in the path is encountered during the traversal, it is determined that there is a cycle. Remove the edge with the smallest weight in the cycle until DFS can no longer detect any cycles and obtains an acyclic consensus matrix. S63: Set a threshold β, and use the threshold cutoff of the acyclic consensus matrix to obtain the final causal structure between disease risk factors.

8. A system for extracting causal relationships of medical risk factors based on dual-constraint guidance, characterized in that, The system includes a memory and a processor; the memory is used to store an application program; the processor is used to run the application program and execute a method for extracting causal relationships of medical risk factors based on dual-constraint guidance as described in any one of claims 1 to 7.

9. A computer storage medium, characterized in that, The computer storage medium stores a remote monitoring program, which, when executed by a processor, implements a method for extracting causal relationships of medical risk factors based on dual-constraint guidance as described in any one of claims 1 to 7.

Images