Evidence-based medicine reasoning method and system based on marginal causal prior knowledge constraints

Abstract

The present invention discloses an evidence-based medical reasoning method and system based on marginal causal prior knowledge constraints, comprising: obtaining risk factors related to the disease to be analyzed and prior knowledge related to the risk factors; constructing a completely undirected graph with the risk factors as nodes, and using the prior knowledge as constraints to determine the marginal causal relationships between the risk factors, thereby constructing a causal network graph, thereby obtaining evidence-based medical reasoning results for the disease to be analyzed. The marginal causal prior is used as a constraint in the causal network structure, and the marginal causal prior relationship is combined with a data-based conditional independence test to determine the marginal causal relationship of several disease-related risk factors, thereby determining factors directly related to the disease and intermediate variables that are related to the disease but do not have a direct causal relationship.

Description

Technical Field

[0001] The present invention relates to the field of evidence-based medical reasoning technology, and in particular to an evidence-based medical reasoning method and system based on marginal causal prior knowledge constraints. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] According to a common definition, a chronic disease is "a condition that persists for one year or longer and requires ongoing medical care, limits daily activities, or both." Examples include typical chronic diseases such as lung cancer, stomach cancer, liver cancer, colorectal cancer, esophageal cancer, hypertension, stroke, coronary heart disease, and diabetes. Each disease is multifactorial, and analyzing the relationships between risk factors is crucial and clinically relevant. Correlation studies of risk factors can uncover previously unrecognized pathologies and provide a degree of theoretical support for uncovering pathologies.

[0004] Current research on the interrelationships between risk factors can only reflect correlations between risk factors, not directly their causal relationships. The gold standard for proving a causal relationship between risk factors and disease is the randomized controlled trial (RCT). However, the results of RCTs only provide a marginal causal relationship and cannot unlock the pathological black box between risk factors and disease, nor can they reveal the mediating factors through which risk factors ultimately influence disease development. Therefore, there is an urgent need to conduct research on the causal relationships between disease risk factors, thereby revealing the pathological and physiological causal relationships between diseases. This has important theoretical significance and clinical guidance for disease prevention and treatment.

[0005] Currently, correlation analysis between risk factors based on Bayesian networks can be used, but it is limited to the analysis of association relationships and lacks causal evidence. Existing Bayesian network structure learning methods include: (1) constraint-based methods, which use conditional independence tests on node pairs to construct network structures; (2) score-based methods, which optimize the search based on a score function; and (3) hybrid methods that combine the first two methods.

[0006] Constraint-based learning methods can describe the Markov equivalent class of directed acyclic graphs (DAGs) of Bayesian network structures using completely partially directed acyclic graphs (CPDAGs). Based on the definition of inferred causal relationships, the results of conditional independence tests (CI) can statistically represent the causal relationships under the hypothesis. However, in practice, CI tests require a large number of samples and are sensitive to noise. If errors in CI tests are caused by insufficient or noisy data, they may lead to more serious errors in the subsequent learning process, resulting in poor performance.

[0007] Therefore, considering various types of prior knowledge is important to reduce the number of CI tests and improve performance. In previous studies, it was assumed that each prior knowledge represents a specific directed edge, such as assuming that the prior knowledge of variable X→variable Y means a directed arc from X to Y in the DAG;

[0008] However, expert knowledge cannot prove whether the prior knowledge provided is a marginal causal relationship. The causal relationship between the two nodes represented by the prior knowledge may contain intermediary nodes. In other words, there is not necessarily a direct causal relationship between X and Y. This means that the causal relationship represented by the prior knowledge may not be a specific directed edge, but a directed path.

[0009] If one wants to find a DAG with marginal prior knowledge, one must select a DAG that satisfies the prior relationship in the equivalence class after the CPDAG is found by the structural learning method. This process can be very inefficient and time-consuming when dozens of variables are involved because the CPDAG must first be learned using a classic constraint-based structural learning method. However, no previous research has linked the CI test in the constraint-based method with marginal causal prior knowledge.

[0010] The core idea of ​​score-based methods is to start from an initial state and update the state through multiple operations so that the score of the new state gradually approaches the optimal score. Due to the use of greedy strategies, these methods are prone to falling into local optimality and cannot explain the scores in terms of causal relationships. In addition, almost all mainstream score-based optimal structure learning methods have a basic assumption that the scoring function is decomposable. This limits the ability to integrate marginal causal prior knowledge. There are few structure learning methods that use EC trees, A* search, and dynamic programming to integrate marginal priors, but compared with other methods, these methods are less efficient and are usually limited to DAGs of about 50 variables, making it difficult to represent large causal networks.

[0011] From this we can see that current constraint-based structure learning methods cannot handle prior knowledge that contains intermediary or mixed causal knowledge, and the learned structure is quite different from the real structure. Summary of the Invention

[0012] In order to solve the above problems, the present invention proposes an evidence-based medical reasoning method and system based on marginal causal prior knowledge constraints, which uses marginal causal priors as constraints in the causal network structure, combines marginal causal prior relationships with data-based conditional independence tests, and judges marginal causal relationships for several disease-related risk factors, thereby determining factors directly related to the disease and intermediate variables that are related to the disease but do not have a direct causal relationship.

[0013] In order to achieve the above object, the present invention adopts the following technical solutions:

[0014] In a first aspect, the present invention provides an evidence-based medical reasoning method based on marginal causal prior knowledge constraints, comprising:

[0015] Obtain risk factors related to the disease to be analyzed and prior knowledge related to risk factors;

[0016] A completely undirected graph is constructed with risk factors as nodes. Using prior knowledge as constraints, the marginal causal relationships between risk factors are determined, thereby constructing a causal network diagram. Specifically, this includes:

[0017] When the condition set is set to zero, the marginal independence between adjacent nodes in the completely undirected graph is tested according to the conditional independence, and the edges between adjacent nodes without causal relationship are deleted;

[0018] Add a condition set to determine whether there is prior knowledge between the two nodes of the edge; if there is no prior knowledge, determine whether there is a separating set in the node set excluding the two nodes that satisfies marginal independence for the two nodes. If so, delete the edge; if there is prior knowledge, determine whether there is a minimum separating set in the node set that satisfies marginal independence for the two nodes. If so, delete the edge;

[0019] At this point, the causal network skeleton is determined, the undirected edges of the causal network skeleton are directed, and the marginal causal relationship between risk factors is determined, so as to obtain the evidence-based medical reasoning results of the disease to be analyzed.

[0020] As an optional implementation, satisfying marginal independence means that: the two nodes X on the opposite edge i and X j , determine whether there is a node in the set of nodes other than the two nodes that makes X i ⊥X j |S holds for the separation set, X i ⊥X j |S means under S condition, X i Independent of X j ,S is (X i ,X j ) is a set that does not contain X i and X j variables.

[0021] As an alternative embodiment, if there exists a separation set S that is adj(C,X i )\{X j} or adj(C,X j )\{X i}, then delete the edge, where the size of the separation set is the size of the conditional set.

[0022] As an alternative implementation, the two nodes X on the opposite edgei and X j , if there is marginal prior knowledge Judge adj(C,X j )\{X i} whether there is such a i ⊥X j |S is the minimum separating set S for which S is established. If it exists, delete the edge.

[0023] As an optional implementation, the size l of the condition set starts from 1. When l=1, after all adjacent nodes are verified, l increases by 1 until the number of adjacent sets of nodes in the completely undirected graph is less than l, thereby determining the causal network skeleton.

[0024] As an optional implementation, the undirected edges of the causal network skeleton are directed based on the MEEK theorem and satisfy prior knowledge.

[0025] As an optional implementation manner, risk factors that have a direct causal relationship with the disease to be analyzed, and risk factors that are related to the disease to be analyzed but have no direct causal relationship are determined based on the causal network diagram.

[0026] In a second aspect, the present invention provides an evidence-based medical reasoning system based on marginal causal prior knowledge constraints, comprising:

[0027] an acquisition module configured to acquire risk factors related to the disease to be analyzed and prior knowledge related to the risk factors;

[0028] The causal network diagram construction module is configured to construct a completely undirected graph with risk factors as nodes, and to determine the marginal causal relationship between risk factors based on prior knowledge constraints, thereby constructing a causal network diagram. Specifically, it includes:

[0029] When the condition set is set to zero, the marginal independence between adjacent nodes in the completely undirected graph is tested according to the conditional independence, and the edges between adjacent nodes without causal relationship are deleted;

[0030] Add a condition set to determine whether there is prior knowledge between the two nodes of the edge; if there is no prior knowledge, determine whether there is a separating set in the node set excluding the two nodes that satisfies marginal independence for the two nodes. If so, delete the edge; if there is prior knowledge, determine whether there is a minimum separating set in the node set that satisfies marginal independence for the two nodes. If so, delete the edge;

[0031] The analysis module is configured to determine the causal network skeleton, orient the undirected edges of the causal network skeleton, and determine the marginal causal relationship between risk factors, so as to obtain the evidence-based medical reasoning results of the disease to be analyzed.

[0032] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0033] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention proposes an evidence-based medical reasoning method and system based on marginal causal prior knowledge constraints. It considers the constraint-based Bayesian network structure learning problem when the prior knowledge is in the form of marginal causal prior knowledge, and proposes a Bayesian network structure learning method based on marginal causal prior knowledge constraints to judge the marginal causal relationship of several disease-related risk factors, thereby determining factors directly related to the disease and intermediate variables that are related to the disease but do not have a direct causal relationship.

[0036] The present invention proposes an evidence-based medical reasoning method and system based on marginal causal prior knowledge constraints, incorporates marginal prior knowledge into a constraint-based Bayesian network structure learning algorithm, uses marginal causal priors as constraints in the causal network structure, combines marginal causal prior relationships with data-based conditional independence tests, and combines statistical hypothesis test results with the causal properties of marginal causal knowledge in the process of network structure learning to obtain more additional causal relationships, further accelerate the efficiency of structure learning, and improve the accuracy of structure learning.

[0037] The present invention proposes an evidence-based medical reasoning method and system based on marginal causal prior knowledge constraints. Under the constraints of marginal causal prior knowledge, the marginal causal relationship between the determined risk factors and the constructed causal network diagram are significantly improved in accuracy. The obtained network structure is closer to the real network structure and has higher performance and efficiency, and can be used to process large network structures.

[0038] A Bayesian network is a graphical network that describes the causal relationships between a large number of variables. The causal relationships between risk factors and diseases obtained from randomized controlled trials can provide more knowledge and information for the study of causal networks between multiple variables. The present invention uses the marginal causal relationships of known randomized controlled trials and observational research data to perform structural learning on the causal relationships between various variables in the causal network, obtaining a more in-depth causal pathway than the results of randomized controlled trials, such as the direct causal relationship in which risk factors affect diseases through various mediating factors, opening the "black box" of marginal causal relationships in randomized controlled trials. In addition, the present invention can also provide more unknown causal knowledge, discover causal relationships that have not been found in existing randomized controlled trials, and provide guidance for further and more detailed new randomized controlled trials and clinical decision-making.

[0039] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0041] Figure 1 A flow chart of the evidence-based medicine reasoning method based on marginal causal prior knowledge constraints provided in Example 1 of the present invention;

[0042] Figure 2 This is an example diagram of the causal network diagram construction process provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0044] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0046] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0047] Example 1

[0048] This embodiment studies the problem of combining marginal causal prior knowledge with a constraint-based Bayesian network structure learning method under the assumption that there are no hidden variables and no selection bias; that is, given a causal marginal prior knowledge that may have multiple directional paths, the CI test is used to discover the statistical characteristics of the prior knowledge and improve the performance and efficiency of network structure learning, so that it can cope with the structure learning of large networks with thousands of variables, with higher computational efficiency, less computing time and better accuracy.

[0049] Based on the above analysis, this embodiment provides an evidence-based medical reasoning method based on marginal causal prior knowledge constraints, such as Figure 1 Shown, including:

[0050] Obtain risk factors related to the disease to be analyzed and prior knowledge related to risk factors;

[0051] A completely undirected graph is constructed with risk factors as nodes. Using prior knowledge as constraints, the marginal causal relationships between risk factors are determined, thereby constructing a causal network diagram. Specifically, this includes:

[0052] When the condition set is set to zero, the marginal independence between adjacent nodes in a completely undirected graph is tested according to conditional independence, and the edges between adjacent nodes without causal relationship are deleted;

[0053] Add a condition set to determine whether there is prior knowledge between the two nodes of the edge; if there is no prior knowledge, determine whether there is a separating set in the node set excluding the two nodes that satisfies marginal independence for the two nodes. If so, delete the edge; if there is prior knowledge, determine whether there is a minimum separating set in the node set that satisfies marginal independence for the two nodes. If so, delete the edge;

[0054] At this point, the causal network skeleton is determined, the undirected edges of the causal network skeleton are directed, and the marginal causal relationship between risk factors is determined, so as to obtain the evidence-based medical reasoning results of the disease to be analyzed.

[0055] In this embodiment, first, the risk factors X related to the disease to be analyzed are obtained, and a completely undirected graph C is constructed with each risk factor as a node;

[0056] Then, we obtain prior knowledge of relevant risk factors from comprehensive disease prevention and control guidelines and articles published in domestic and foreign journals. Based on the prior knowledge, we determine the marginal causal relationship between risk factors and construct a causal network diagram of the disease to be analyzed.

[0057] In this embodiment, the input is the observed data of the disease and marginal prior knowledge; in fact, this embodiment also sets up input interfaces for direct prior knowledge and marginal prior knowledge. However, all prior knowledge must be regarded as marginal prior knowledge because in the absence of intermediate variables, it is impossible to determine whether one variable affects another variable.

[0058] In this embodiment, first, when the condition set size l = 0, that is, there is no intermediate variable between two adjacent nodes, the marginal independence of all variable pairs (i.e., adjacent nodes) in the completely undirected graph is tested using conditional independence to determine whether the variable pair X i 、X j Is there a causal relationship between X and i ⊥X j , then X i Independent of X j , which is X i 、X j There is no causal relationship between them, so delete edge X from the completely undirected graph C. i -X j , and obtain the updated first completely undirected graph C1.

[0059] Then, since marginal prior knowledge can only extend more causal knowledge when there are variables in the condition set, increasing the size of the condition set requires additional testing when l ≥ 1;

[0060] Specifically: when l=1, select the adjacent ordered vertex pairs (X i ,X j ), judge X i and X j Is there marginal prior knowledge between them?

[0061] (1) If there is no marginal prior knowledge, according to the conditional independence test, it is determined that except for the two nodes X i and X j Is there a node in the set of nodes outside X?i ⊥X j |S holds true for a separation set S; that is, whether there is a separation set S that satisfies X i 、X j Separating set S of marginal independence conditions;

[0062] Where S is (X i ,X j ) is a set that does not contain X i and X j Variables (can also be considered as condition sets); X i ⊥X j |S holds true means that under the S condition, X i Independent of X j , then we need to delete edge X i -X j , can also be understood as, in X i 、X j Add intermediate variables to judge X i ⊥X j |S.

[0063] When l = 1, the separation set S is adj(C,X i )\{X j} or a subset of size l = 1 in adj(C,X j )\{X i} is a subset of size l = 1, because adj(C,X j ) or any subset of adj(C,X i ) may be any subset of X i and X j Separating set of adj(C,X j ) is X in network C j The set of all adjacent nodes of adj(C,X j )\{X i} is except X i Node X in the network C outside j The set of all other adjacent nodes of

[0064] Then, for X i Not X j The marginal prior knowledge of the ancestors, if in adj(C,X i )\{X j There is always a separating set S in the subset of}, or for X j Not X i The marginal prior knowledge of the ancestors, if in adj(C,X j )\{X iThere is always a separating set S in the subset of X such that i 、X j Satisfy the marginal independence condition, which means X i 、X j If there is no direct causal relationship, the edge is deleted;

[0065] If no subset satisfies the independence condition, then X i With X j There is a direct causal relationship and the edge is retained.

[0066] (2) If there is marginal prior knowledge This means that X i It's X j An ancestor of j )\{X i} whether there is such a i ⊥X j |The minimum separating set S for which S is established, the size of S is l = 1, if it exists, delete the edge;

[0067] Equivalent to, given X j The parent node, X i With X j No relation, X j The parent node must be the same as X j Adjacent, therefore, we can always find a minimum separation set S that is adj(C,X j )\{X i}.

[0068] In addition, if based on adj(C,X i )\{X j} Determine whether there is a subset. It is possible that no separation set can be found; this is because, given a variable X i Independent of its child nodes and their parent nodes, which may be different from X i are not adjacent, so it is possible that adj(C,X i )\{X j None of the subsets of} is X i and X j Separation set.

[0069] In summary, in this embodiment, adj(C,X i )\{X j No need to search for X in} i and X j The separation set S can reduce the number of independent tests and only needs to find X i and X j The minimum separating set is sufficient.

[0070] The above is for the case of l=1. If all adjacent nodes have been verified under l=1, then l is increased by 1 and the above process is repeated until the number of adjacent sets of nodes in the completely undirected graph is less than l; at this point, the causal network skeleton and separation set are determined.

[0071] In this embodiment, the causal network skeleton is determined under given marginal prior knowledge, and then a V-shaped structure is found based on this. The orientation is performed based on the MEEK theorem and satisfies the marginal prior knowledge. The orientation principle is as follows:

[0072] (1) If there is a directed edge X i →X j , X j -X k Oriented to X j →X k , so that X i and X k Not adjacent (otherwise a new V-shaped structure will appear);

[0073] (2) If there is a chain X i →X k →X j , then X i -X j Oriented to X i →X j (Otherwise there will be a directed cycle);

[0074] (3) When there are two chains X i -X k →X j and X i -X l →X j , so that X k and X l When not adjacent, X i -X j Oriented to X i →X j (otherwise a new V-shaped structure or directed cycle will appear); such as Figure 2 (a)-(e) show example diagrams of the causal network diagram construction process.

[0075] In this embodiment, taking liver cancer as an example, the process of the evidence-based medicine reasoning method for liver cancer is as follows:

[0076] Obtain liver cancer-related risk factors from the regional health checkup cohort data of the Health and Medical Big Data Research Institute Collaborative Center;

[0077] The cohort data includes 2080 standard indicator variables, including demographic information (age, gender, etc.), disease (ICD-10 code), drug (YPID code), laboratory test (CNAS-AL09), etc.

[0078] Risk factors associated with liver cancer include 1,468 variables, including blood indicators, gender, age, height, weight, pulse, diastolic blood pressure, systolic blood pressure, body mass index, and diseases;

[0079] Prior information on the aforementioned liver cancer risk factors was obtained from comprehensive disease prevention and control guidelines and articles published in domestic and international journals. This known prior information was used to construct marginal causal relationships between the variables included in the prior information. A causal network diagram for liver cancer was constructed using a Bayesian network structure learning method constrained by marginal causal prior knowledge.

[0080] The causal network diagram includes direct causal variables that have a direct causal relationship with liver cancer, and intermediate variables that are related to liver cancer but have no direct causal relationship;

[0081] Therefore, the direct causal variables of liver cancer are: γ-glutamyl transpeptidase, aspartate aminotransferase, carcinoembryonic antigen, liver fibrosis and cirrhosis, gallbladder, biliary tract and pancreatic diseases, other diseases of the digestive system, non-insulin-dependent diabetes mellitus, gender, cholecystitis, other diseases of the bile duct, and other diseases of the pancreas, which are consistent with the existing disease prevention and control guidelines, expert experience and the conclusions of published articles, indicating the effectiveness of this method.

[0082] The Bayesian network structure learning method based on marginal causal prior knowledge constraints proposed in this embodiment has higher accuracy, recall rate, F1 score and minimum structural distance, can handle large-scale network structure learning tasks of 1,000 nodes, has faster computing speed and requires less computational effort.

[0083] Example 2

[0084] This embodiment provides an evidence-based medical reasoning system based on marginal causal prior knowledge constraints, including:

[0085] an acquisition module configured to acquire risk factors related to the disease to be analyzed and prior knowledge related to the risk factors;

[0086] The causal network diagram construction module is configured to construct a completely undirected graph with risk factors as nodes, and to determine the marginal causal relationship between risk factors based on prior knowledge constraints, thereby constructing a causal network diagram. Specifically, it includes:

[0087] When the condition set is set to zero, the marginal independence between adjacent nodes in the completely undirected graph is tested according to the conditional independence, and the edges between adjacent nodes without causal relationship are deleted;

[0088] Add a condition set to determine whether there is prior knowledge between the two nodes of the edge; if there is no prior knowledge, determine whether there is a separating set in the node set excluding the two nodes that satisfies marginal independence for the two nodes. If so, delete the edge; if there is prior knowledge, determine whether there is a minimum separating set in the node set that satisfies marginal independence for the two nodes. If so, delete the edge;

[0089] The analysis module is configured to determine the causal network skeleton, orient the undirected edges of the causal network skeleton, and determine the marginal causal relationship between risk factors, so as to obtain the evidence-based medical reasoning results of the disease to be analyzed.

[0090] In this embodiment, satisfying marginal independence means that: the two nodes X on the opposite edge i and X j , determine whether there is a node in the set of nodes other than the two nodes that makes X i ⊥X j |S holds for the separation set, X i ⊥X j |S means under S condition, X i Independent of X j ,S is (X i ,X j ) is a set that does not contain X i and X j variables.

[0091] In this embodiment, if there is a separation set S that is adj(C,X i )\{X j} or adj(C,X j )\{X i}, then delete the edge, where the size of the separation set is the size of the conditional set.

[0092] In this embodiment, the two nodes X on the opposite side i and X j , if there is marginal prior knowledge Judge adj(C,X j )\{X i} whether there is such a i ⊥X j |S is the minimum separating set S for which S exists, and if so, delete the edge.

[0093] In this embodiment, the size l of the condition set starts from 1. When l=1, after all adjacent nodes are verified, l increases by 1 until the number of adjacent sets of nodes in the completely undirected graph is less than l, and the causal network skeleton is determined.

[0094] In this embodiment, the undirected edges of the causal network skeleton are directed based on the MEEK theorem and satisfy prior knowledge.

[0095] In this embodiment, risk factors that have a direct causal relationship with the disease to be analyzed and risk factors that are related to the disease to be analyzed but have no direct causal relationship are determined based on the causal network diagram.

[0096] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0097] In further embodiments, there is also provided:

[0098] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor, wherein when the computer instructions are executed by the processor, the method described in Example 1 is performed. For the sake of brevity, no further details are given here.

[0099] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0100] The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0101] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is performed.

[0102] The method in Example 1 can be directly implemented as a hardware processor, or can be implemented using a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, it will not be described in detail here.

[0103] Those skilled in the art will appreciate that the units, i.e., algorithm steps, of the various examples described in conjunction with this embodiment can be implemented using electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0104] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. An evidence-based medical reasoning method based on marginal causal prior knowledge constraints, characterized by: include: Obtain risk factors related to the disease to be analyzed and prior knowledge related to risk factors; A completely undirected graph is constructed with risk factors as nodes. Using prior knowledge as constraints, the marginal causal relationships between risk factors are determined, thereby constructing a causal network diagram. Specifically, this includes: When the condition set is set to zero, the marginal independence between adjacent nodes in the completely undirected graph is tested according to the conditional independence, and the edges between adjacent nodes without causal relationship are deleted; Add a condition set to determine whether there is prior knowledge between the two nodes of the edge; if there is no prior knowledge, determine whether there is a separating set in the node set excluding the two nodes that satisfies marginal independence for the two nodes. If so, delete the edge; if there is prior knowledge, determine whether there is a minimum separating set in the node set that satisfies marginal independence for the two nodes. If so, delete the edge; At this point, the causal network skeleton is determined, the undirected edges of the causal network skeleton are directed, and the marginal causal relationship between risk factors is determined, so as to obtain the evidence-based medical reasoning results of the disease to be analyzed.

2. The evidence-based medical reasoning method based on marginal causal prior knowledge constraints according to claim 1 is characterized in that: Satisfying marginal independence means: the two nodes on the opposite edge X i and X j , to determine whether there is a node in the node set other than the two nodes. X i ⊥ X j |S holds for the separation set, X i ⊥ X j |S means under S condition, X i Independent of X j , S is A disjunctive set is a set that does not contain X i and X j variables.

3. The evidence-based medical reasoning method based on marginal causal prior knowledge constraints according to claim 2 is characterized in that: If there exists a separating set S that is A subset of If there is a subset of , then the edge is deleted, where the size of the separation set is the size of the condition set, For network C The set of all adjacent nodes of For the exception In the network C outside the node The set of all other adjacent nodes.

4. The evidence-based medical reasoning method based on marginal causal prior knowledge constraints as claimed in claim 3, characterized in that: Two nodes on opposite sides X i and X j , if there is marginal prior knowledge X i ⇝ X j ,judge Is there a X i ⊥ X j |S is the minimum separating set S for which S exists, and if so, delete the edge.

5. The evidence-based medical reasoning method based on marginal causal prior knowledge constraints according to claim 1 is characterized in that: Size of the condition set l Starting from 1, l= 1, after all adjacent nodes have been verified, l Increase by 1 until the number of adjacent sets of nodes in the completely undirected graph is less than l , thus determining the causal network skeleton.

6. The evidence-based medical reasoning method based on marginal causal prior knowledge constraints according to claim 1 is characterized in that: The undirected edges of the causal network skeleton are directed based on the MEEK theorem and satisfy prior knowledge.

7. The evidence-based medical reasoning method based on marginal causal prior knowledge constraints according to claim 1 is characterized in that: Based on the causal network diagram, the risk factors that have a direct causal relationship with the disease to be analyzed and the risk factors that are related to the disease to be analyzed but have no direct causal relationship are determined.

8. An evidence-based medical reasoning system based on marginal causal prior knowledge constraints, characterized by: include: an acquisition module configured to acquire risk factors related to the disease to be analyzed and prior knowledge related to the risk factors; The causal network diagram construction module is configured to construct a completely undirected graph with risk factors as nodes, and to determine the marginal causal relationship between risk factors based on prior knowledge constraints, thereby constructing a causal network diagram. Specifically, it includes: When the condition set is set to zero, the marginal independence between adjacent nodes in the completely undirected graph is tested according to the conditional independence, and the edges between adjacent nodes without causal relationship are deleted; Add a condition set to determine whether there is prior knowledge between the two nodes of the edge; if there is no prior knowledge, determine whether there is a separating set in the node set excluding the two nodes that satisfies marginal independence for the two nodes. If so, delete the edge; if there is prior knowledge, determine whether there is a minimum separating set in the node set that satisfies marginal independence for the two nodes. If so, delete the edge; The analysis module is configured to determine the causal network skeleton, orient the undirected edges of the causal network skeleton, and determine the marginal causal relationship between risk factors, so as to obtain the evidence-based medical reasoning results of the disease to be analyzed.

9. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 7 is completed.

10. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the method according to any one of claims 1 to 7.

Images