A traditional Chinese medicine clinical diagnosis and treatment model construction method and clinical auxiliary decision method based on a clinical practice guideline

Abstract

The application relates to the technical field of computer application, and discloses a traditional Chinese medicine clinical diagnosis and treatment model construction method and a clinical auxiliary decision-making method based on a clinical practice guideline, wherein the traditional Chinese medicine clinical diagnosis and treatment model construction method fuses a medical knowledge base and a preset language model to construct a traditional Chinese medicine clinical diagnosis and treatment model with a cause-effect relationship, effectively overcoming the problems of incomplete knowledge coverage, lagging update and subjective bias caused by the dependence of traditional methods on expert experience; the clinical auxiliary decision-making method dynamically adjusts the model by introducing clinical diagnosis and treatment scenes and context information, can effectively inhibit the interference of irrelevant nodes, enhance the contribution of nodes and directed edges matched with the current case, and make the model more suitable for actual clinical situations; and the initial state vector and the effective reasoning subgraph are combined for iterative reasoning, improving the pertinence and accuracy of diagnosis.

Description

Technical Field

[0001] This invention relates to the field of computer application technology, specifically to a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and a clinical decision support method. Background Technology

[0002] Clinical practice guidelines integrate high-quality evidence to standardize clinical assessment, diagnosis, and treatment decisions. However, they are mostly presented as narrative texts, lacking calculable structured representations and quantitative analysis of relationship strengths, which limits their support for multi-step, individualized decision-making across various diagnostic and treatment stages in real-world clinical scenarios. Previous studies have proposed various decision support tools based on clinical practice guidelines and medical decision support methods based on Fuzzy Cognitive Maps (FCMs) to improve the standardization and consistency of diagnosis and treatment. Among these, the relevant technologies mainly include: medical FCM decision support systems built primarily based on expert experience, combined with limited guideline or literature information. These systems typically involve clinical experts listing conceptual nodes such as symptoms, signs, examination indicators, diseases, and treatment plans based on experience. Through expert discussion or the Delphi method, the causal edges and weights between each node are subjectively set, forming an FCM model for a specific disease or disease type. In clinical use, the system assigns initial activation values ​​to relevant nodes based on the patient's current symptoms and examination results, calculates the activation degree of other nodes through iterative propagation, and finally outputs several recommended disease diagnoses or treatment plans for physicians' reference.

[0003] The aforementioned closest related technologies have, to some extent, achieved the goal of formalizing and quantifying expert knowledge, and can provide auxiliary diagnostic or treatment suggestions for specific diseases within a limited scope. However, they still have the following defects and shortcomings: (1) The knowledge sources rely on the experience of a few experts and fail to systematically utilize clinical practice guidelines, resulting in insufficient knowledge coverage and difficulty in updating. The nodes and edges of the existing FCM decision support system are mostly manually compiled by a few experts based on personal experience and limited literature, lacking a systematic knowledge extraction and integration mechanism based on authoritative clinical practice guidelines. As a result, the model does not fully cover the multi-level diagnostic and treatment elements in the field of liver disease, such as "symptoms-syndromes-diseases-treatment principles-prescriptions-traditional Chinese medicines", and some guideline recommendations cannot be reflected; when the guidelines are updated or new expert consensus is released, the model is difficult to synchronize in a timely manner, requiring manual sorting and modification of the graph structure, resulting in high maintenance costs; there are large differences between FCM models built by different experts or different teams, lacking a unified knowledge base, which is not conducive to promotion and application.

[0004] (2) Edge weights rely entirely or primarily on manual assignment, lacking unified quantitative standards and prone to subjective bias. In related technologies, the causal strength and direction between concept nodes are usually given through expert scoring or empirical estimation, lacking reliable automated or semi-automated quantitative methods. Different experts have different understandings of the importance of the same relationship, and the scoring standards are inconsistent, resulting in strong subjectivity in edge weights. As the number of nodes and edges increases, it is difficult to ensure overall consistency and maintainability by relying solely on manual assignment. Large-scale text data and existing model knowledge are rarely used to objectively estimate the strength of relationships. This type of subjective weighting method easily leads to an over-reliance on the preferences of a few experts in FCM inference results, making it difficult to reflect the consensus patterns implied in large-sample guidelines and literature.

[0005] (3) The model structure and reasoning mechanism are difficult to support multi-level and multi-task liver disease diagnosis and treatment decisions simultaneously, and the functions are relatively simple. Existing FCM decision support systems are mostly geared towards a single level or a few tasks, such as only facing the diagnostic reasoning of "symptom → disease" or only recommending a certain type of treatment plan. When it comes to liver disease and traditional Chinese medicine liver disease, this design is difficult to cover the multi-level decision-making process that is actually needed in clinical practice: it is difficult to handle chain reasoning such as "symptom → syndrome", "syndrome → disease" and "syndrome → treatment principle / prescription / traditional Chinese medicine" in the same model at the same time; it is not convenient to give a comprehensive recommendation of diagnosis, syndrome identification and treatment plan in the same reasoning process, and doctors cannot examine the entire diagnosis and treatment path in a unified framework; when trying to expand to more node types and tasks, the structural design and numerical stability problems of traditional FCM become more prominent. Summary of the Invention

[0006] This invention provides a method for constructing a TCM clinical diagnosis and treatment model and a clinical decision support method based on clinical practice guidelines, in order to solve the technical problems of existing disease decision support models being highly subjective, having an incomplete knowledge system, and lacking diagnostic reasoning ability.

[0007] In a first aspect, the present invention provides a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines, comprising the following steps: extracting semantic features representing disease symptoms, syndromes, and treatment plans from a medical knowledge base; constructing a network topology containing nodes and directed edges between nodes, wherein the nodes include symptom nodes, syndrome nodes, and treatment nodes; determining the causal strength values ​​of the directed edges in the network topology using a preset language model to obtain a weight matrix; defining a weight membership function for the directed edges to map the causal strength values ​​of the directed edges to causal strength levels; and generating a TCM clinical diagnosis and treatment model based on the network topology, the weight matrix, and the weight membership function.

[0008] The present invention provides a method for constructing a TCM clinical diagnosis and treatment model, proposing a novel reasoning mechanism GM-FCM composed of a unified multi-step task framework, stable dynamic propagation, first-hop input gating, syndrome context modulation, intra-group competition normalization, and structural prior constraints. By integrating a medical knowledge base and a pre-set language model, a TCM clinical diagnosis and treatment model with causal relationships is constructed, effectively overcoming the problems of incomplete knowledge coverage, delayed updates, and subjective biases caused by the reliance on expert experience in traditional methods. By using a language model to quantify the causal strength between nodes and defining a weighted membership function, the model achieves objective and refined modeling of the correlation between diagnostic and treatment elements, improving the model's ability to express the complex multi-level relationship of "symptom-syndrome-treatment". The generated diagnosis and treatment model can support multi-task joint reasoning, providing integrated decision support for disease diagnosis, syndrome identification, and treatment plan recommendation.

[0009] In some optional implementations, the causal strength values ​​of directed edges in the network topology are determined using a preset language model to obtain a weight matrix, which includes: evaluating the causal relationship strength of the same directed edge using multiple language models to obtain multiple initial evaluation weights for each directed edge; obtaining the credibility score of each language model and determining the weight coefficient of each language model based on the credibility score; and weighting and fusing the multiple initial evaluation weights belonging to the same directed edge based on the weight coefficient of each language model to obtain the causal strength value of each directed edge.

[0010] This implementation method, by introducing a multilingual model collaborative evaluation mechanism and combining it with a model credibility weighted fusion strategy, can effectively avoid the bias and uncertainty caused by single model evaluation, and significantly improve the objectivity and reliability of causal strength values.

[0011] In some optional implementations, the weighted membership function is a fuzzy membership function, used to achieve a fuzzy mapping from causal strength values ​​to causal strength levels.

[0012] This implementation uses a fuzzy membership function to map continuous causal intensity values ​​into discrete causal intensity levels, which can effectively handle the inherent fuzziness and uncertainty in disease diagnosis and treatment knowledge, making the model closer to clinical practice. At the same time, the fuzzy mapping mechanism avoids the information loss caused by traditional binary judgment and improves the granularity and rationality of the reasoning results.

[0013] Secondly, the present invention also provides a clinical decision support method, comprising the following steps: acquiring clinical information of a patient; wherein the clinical information includes clinical diagnosis and treatment data, clinical diagnosis and treatment scenarios, and clinical context information; mapping the clinical information to a traditional Chinese medicine (TCM) clinical diagnosis and treatment model constructed using the first aspect or any of its corresponding implementation methods; adjusting the node activation states and directed edge weights in the TCM clinical diagnosis and treatment model according to the clinical diagnosis and treatment scenarios and clinical context information, so as to suppress nodes that do not match the clinical diagnosis and treatment scenarios and clinical context information, and enhance nodes and directed edges that match the clinical diagnosis and treatment scenarios and clinical context information, thereby obtaining an initial state vector and an effective inference subgraph; and performing iterative inference on the TCM clinical diagnosis and treatment model based on the initial state vector, the effective inference subgraph, and the clinical diagnosis and treatment data, until the state of the TCM clinical diagnosis and treatment model converges, thereby obtaining the inference result.

[0014] The clinical auxiliary decision-making system based on the TCM clinical diagnosis and treatment model provided by this invention can effectively suppress the interference of irrelevant nodes and enhance the contribution of nodes and directed edges that match the current case by introducing clinical diagnosis and treatment scenarios and contextual information to dynamically adjust the model, making the model more in line with the actual clinical situation. Iterative reasoning by combining the initial state vector and the effective reasoning subgraph improves the pertinence and accuracy of diagnosis.

[0015] In some optional implementations, the TCM clinical diagnosis and treatment model is iteratively reasoned based on the initial state vector, the effective inference subgraph, and clinical diagnosis and treatment data until the state of the TCM clinical diagnosis and treatment model converges, and the reasoning result is obtained. This includes: using the initial state vector as the initial input for iterative reasoning, in each iteration of the TCM clinical diagnosis and treatment model, calculating the weighted input value of the node in the current round based on the directed edge weights and node activation states in the effective inference subgraph; using the hyperbolic tangent function with zero center as the node activation function to map and calculate the weighted input value to obtain the intermediate activation value of the node in the current round; obtaining the final activation value of the node in the previous round; weighting and fusing the intermediate activation value of the node in the current round with the final activation value of the node in the previous round according to a preset damping coefficient to obtain the final activation value of the node in the current round; and repeating the above iterative process until the TCM clinical diagnosis and treatment model converges to obtain the reasoning result.

[0016] This implementation uses a zero-centered hyperbolic tangent function as the activation function, which improves the model's ability to express positive and negative causal relationships and makes the distribution of activation values ​​more balanced. By introducing a damping coefficient to fuse historical activation states, the stability of the model's inference process is effectively enhanced, avoiding violent oscillations or divergence of node states during iteration and ensuring smooth convergence of the inference process.

[0017] In some optional implementations, after obtaining the reasoning results, the method further includes: dividing the nodes in the TCM clinical diagnosis and treatment model into multiple functional node groups according to their functional attributes, wherein each functional node group includes at least a symptom node group, a syndrome node group, and a treatment node group; for each functional node group, performing an intra-group competitive decision-making operation: calculating the sum of the activation values ​​of all nodes within the functional node group to obtain a normalized benchmark; based on the normalized benchmark, normalizing the activation values ​​of each node within the functional node group to obtain the relative importance of each node; selecting nodes with a relative importance higher than a preset threshold as candidate recommended nodes, and sorting the candidate recommended nodes according to their relative importance to obtain the diagnosis and treatment recommendation results for the functional node group.

[0018] This implementation method introduces an intragroup competitive decision-making mechanism and uses normalization to quantify the relative importance of nodes, which can effectively screen out key diagnostic and treatment elements in each dimension, thereby improving the interpretability of model inference results and the efficiency of clinical decision-making.

[0019] In some optional implementations, after sorting the candidate recommendation nodes according to their relative importance to obtain the diagnosis and treatment recommendation results of the functional node group, the method further includes: selecting high-activation nodes and high-weight edges related to the diagnosis and treatment recommendation results from the convergence state and weight matrix of the TCM clinical diagnosis and treatment model; constructing representative diagnosis and treatment paths based on the high-activation nodes and high-weight edges, generating structured path data corresponding to the representative diagnosis and treatment paths; and generating natural language interpretations based on the structured path data.

[0020] This implementation method generates typical treatment paths by screening highly activated nodes and high-weight edges, and then converts them into natural language explanations, which greatly improves the interpretability and readability of TCM clinical diagnosis and treatment model decisions, making the recommendation basis clear and easy to understand, and more conducive to clinical application and promotion.

[0021] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines as described in the first aspect or any corresponding embodiment above, and / or the clinical auxiliary decision-making method as described in the second aspect or any corresponding embodiment above.

[0022] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to the first aspect or any corresponding embodiment described above, and / or the clinical auxiliary decision-making method according to the second aspect or any corresponding embodiment described above.

[0023] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to the first aspect or any corresponding embodiment described above, and / or the clinical auxiliary decision-making method according to the second aspect or any corresponding embodiment. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a first flowchart of a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to an embodiment of the present invention; Figure 3 This is a second flowchart of the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the semantic feature extraction process according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the network topology according to an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the data mapping between entities and concepts according to an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the causal relationship between concepts according to embodiments of the present invention; Figure 8 This is a schematic diagram of the weight matrix according to an embodiment of the present invention; Figure 9 This is a first flowchart of a clinical decision support method according to an embodiment of the present invention; Figure 10 This is a schematic diagram illustrating the introduction of gating and context modulation mechanisms during the inference phase according to an embodiment of the present invention; Figure 11 This is a second flowchart of a clinical decision support method according to an embodiment of the present invention; Figure 12 This is a schematic diagram illustrating the process of case input, state initialization, iterative update, and result output according to an embodiment of the present invention; Figure 13 This is a schematic diagram of a traditional Chinese medicine clinical diagnosis and treatment model according to an embodiment of the present invention; Figure 14 This is a schematic diagram of a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and a clinical decision support method according to an embodiment of the present invention; Figure 15 This is a schematic diagram illustrating the task-level evaluation of a TCM clinical diagnosis and treatment model and various comparison methods according to an embodiment of the present invention; Figure 16 This is a structural block diagram of a device for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to an embodiment of the present invention; Figure 17 This is a structural block diagram of a clinical decision support device according to an embodiment of the present invention; Figure 18 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0028] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0029] As an optional application scenario of this invention, such as Figure 1 As shown, the construction method of TCM clinical diagnosis and treatment model and the application scenarios of clinical auxiliary decision-making system can include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0030] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0031] According to an embodiment of the present invention, a method for constructing a TCM clinical diagnosis and treatment model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] This embodiment provides a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines, which can be used on the aforementioned mobile terminals, such as mobile phones and tablets. Figure 2 This is a first flowchart of a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process includes the following steps: Step S201: Extract semantic features representing disease symptoms, syndromes and treatment plans from the medical knowledge base, and construct a network topology structure containing nodes and directed edges between nodes, where nodes include symptom nodes, syndrome nodes and treatment nodes.

[0033] Step S202: Use a preset language model to determine the causal strength values ​​of directed edges in the network topology and obtain the weight matrix.

[0034] Step S203: Define the weight membership function of the directed edge to map the causal strength value of the directed edge to the causal strength level.

[0035] Step S204: Generate a TCM clinical diagnosis and treatment model based on the network topology, weight matrix, and weight membership function. The method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines provided in this embodiment integrates a medical knowledge base with a pre-defined language model to build a TCM clinical diagnosis and treatment model with causal relationships. This effectively overcomes the problems of incomplete knowledge coverage, delayed updates, and subjective biases caused by the reliance on expert experience in traditional methods. By using a language model to quantify the causal strength between nodes and defining the weight membership function, the model achieves objective and refined modeling of the correlation between diagnostic and treatment elements, improving its ability to express the complex multi-level relationships of "symptom-syndrome-treatment." The generated diagnosis and treatment model can support multi-task joint reasoning, providing integrated decision support for disease diagnosis, syndrome identification, and treatment plan recommendation.

[0036] This embodiment provides a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines, which can be used on the aforementioned mobile terminals, such as mobile phones and tablets. Figure 3 This is a second flowchart of the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines according to an embodiment of the present invention, as shown below. Figure 3 As shown, the process includes the following steps: Step S301: Extract semantic features representing disease symptoms, syndromes and treatment plans from the medical knowledge base, and construct a network topology structure containing nodes and directed edges between nodes, where nodes include symptom nodes, syndrome nodes and treatment nodes.

[0037] The need for this invention is particularly prominent in the context of liver disease diagnosis and treatment. Liver disease is not a single disease entity, but rather a continuous spectrum covering end-stage manifestations ranging from acute and chronic liver injury to cirrhosis, decompensation, and liver failure. Clinicians need to comprehensively refer to multiple liver disease guidelines at different disease stages or on different topics within a short period of time for diagnosis and treatment. Therefore, liver disease is selected as an example to test the ability of this invention to solve the technical challenges of existing decision support models, such as strong subjectivity, incomplete knowledge systems, and insufficient diagnostic reasoning capabilities.

[0038] Taking liver disease as an example, the purpose of extracting semantic features representing liver disease symptoms, syndromes, and treatment plans from the medical knowledge base is to convert the text data in the medical knowledge base (including liver disease-related clinical practice guidelines, expert consensus, etc.) into standardized inputs that can be used for subsequent structured extraction and computational reasoning, and to establish a unified terminology system, such as... Figure 4 The steps shown include the following: (1) Obtain relevant clinical practice guidelines, expert consensus and normative documents for liver diseases, convert the original documents into machine-readable formats (such as UTF-8 text or documents with structure tags), and retain metadata such as guide source, version number, chapter number, clause number, etc.

[0039] (2) The guide text is denoised and cleaned (the table of contents, headers and footers, copyright notices, duplicate paragraphs, abnormal symbols, etc. are deleted), and it is segmented and numbered at the granularity of chapters / clauses / paragraphs to form a set of structured text units, which provide basic processing units for subsequent entity extraction and relationship recognition.

[0040] (3) Extract diagnosis and treatment-related entities (including symptoms / signs, examination indicators, syndromes, diseases, treatment principles, prescriptions, Chinese medicines, etc.) from structured text units. Entity word lists can be generated by combining dictionary matching, rule templates, and large language model-assisted recognition.

[0041] (4) Perform term alignment and standardization on candidate terms, including synonym merging, alias mapping, ambiguity resolution, granularity unification (such as the standardization of "liver stagnation and spleen deficiency" and "liver stagnation and spleen deficiency syndrome"), and cross-guideline consistency alignment to form a unified standard term library; assign a unique code to each standard term and record its alias, source and corresponding guideline clause information.

[0042] This step transforms guide texts from natural language descriptions into computable, traceable, and scalable standardized knowledge inputs, resolving the issue of conceptual inconsistencies arising from multiple guide sources and different expressions. It lays the foundation for subsequent node definition, edge relation extraction, and weight calculation in GM-FCM.

[0043] The purpose of this step is to mine the semantic relationships in the guide text in order to construct, for example... Figure 5 The topology of GM-FCM is shown below. Specifically, by performing syntactic analysis or template matching on guide paragraphs containing multiple entities, relational patterns such as "hint," "belong to," "cause," "suitable," "adopt," "treat," and "compatible" are identified; the identified relations are mapped to directed edges (Ci, Cj) between entities, and the signs of the edges are determined (e.g., support / promotion is positive, inhibition / rejection is negative); an initial graph structure is generated, containing multiple types of nodes such as symptoms / signs, examination indicators, syndromes, diseases, treatment principles, prescriptions, and traditional Chinese medicines, as well as corresponding directed edges; the graph structure is stored as an adjacency matrix or an equivalent data structure, providing a foundation for subsequent weight calculation and reasoning. For example, Figure 6 This demonstrates the data mapping between entities and concepts. Figure 7 It demonstrates the causal relationships between concepts.

[0044] Step S302: Use a preset language model to determine the causal strength values ​​of directed edges in the network topology and obtain the weight matrix.

[0045] In some optional implementations, the causal strength values ​​of directed edges in the network topology are determined using a preset language model to obtain the weight matrix, including the following steps S3021 to S3023.

[0046] Step S3021: Use multiple language models to evaluate the causal relationship strength of the same directed edge, and obtain multiple initial evaluation weights for each directed edge.

[0047] Step S3022: Obtain the credibility score of each language model, and determine the weight coefficient of each language model based on the credibility score of each language model.

[0048] Step S3023: Based on the weight coefficient of each language model, multiple initial evaluation weights belonging to the same directed edge are weighted and fused to obtain the causal strength value of each directed edge.

[0049] Step S303: Define the weight membership function of the directed edge to map the causal strength value of the directed edge to the causal strength level.

[0050] For example, for each candidate directed edge (C) in GM-FCM i C j To determine edge weights, a multi-language model fusion approach combining large-scale language models and expert input was employed. Firstly, initial weights were assigned to the large-scale language models. To quickly and efficiently obtain the initial causal strength between concept nodes, four advanced pre-trained large-scale language models were selected: GPT-4.1-mini, Claude Sonnet 4, Gemini 2.5 Pro (preview), and DeepSeek (R1). The processed structured input data and the set prompt were submitted to the LLM API, with parameters set to temperature=0.5 and max_tokens=10. The four large-scale language models were then used to evaluate the initial quantitative weights of the causal relationships between each pair of concept nodes. W={ } is the adjacency matrix of FCM. Represents concept node C i For C j The intensity of the fuzzy influence, for any directed arc <C i C j >Order (1) This represents the initial evaluation weight of the k-th large language model for the directed arc. k=1, k=2, k=3, and k=4 represent the four large language models: GPT-4.1-mini, Claude Sonnet 4, Gemini 2.5 Pro (preview), and DeepSeek (R1), respectively.

[0051] Second, expert scoring calibration and normalized weights. Considering the advantages of pre-trained large language models in terms of text coverage and their potential domain bias, this study introduced five experts with ≥20 years of clinical experience in TCM diseases to verify and correct the causal strength predicted by the models. A multi-expert consensus mechanism was used to suppress subjective bias and improve the rigor and credibility of the FCM side weights. Domain experts scored and calibrated the preliminary evaluation results of the four models using a 1-5 Likert scale (5 being the most reliable). The normalized weight coefficients of the five experts' reliability scores for the four models are as follows: (2) By weighted fusion, a directed arc is obtained. <C i C j Final weight: (3) Among them W ij A positive value indicates a positive correlation, W ij =1 represents the strongest positive correlation; a negative value indicates a negative correlation. W ij = -1 represents the strongest negative correlation; zero indicates no causal correlation.

[0052] The fuzzy membership degree of the association weights between disease entities is characterized by a triangular membership function, and its general form is defined as follows.

[0053] (4) in = , where is the normalized weight value, and (a,b,c) are the left, right, and left endpoints of the triangle. For any ∈[ 1,1], we first determine the interval [a,b,c] (positive or negative) in which it falls, and then calculate according to the above formula. ( By adjusting the values ​​of a, b, and c, different fuzzy subsets can be defined and overlapped. Typically, linguistic variables are categorized as low, medium, and high. We map the fuzzy values ​​onto the FCM causal network diagram and visualize the strength of fuzzy relationships between entities using directed edges of different colors. For example... Figure 8 The resulting weighted rectangle is constructed.

[0054] By adopting the above steps, we can make full use of the large language model's ability to understand complex texts, transform the causal strength information implicit in the guide text into quantitative edge weights, and constrain and correct the model performance through expert evaluation, avoiding complete reliance on subjective assignment. At the same time, visualizing the mapping weights improves interpretability.

[0055] While estimating edge weights using only a single large language model without expert weighting and fusion is relatively simple, the results are highly dependent on the version and training status of the single model, resulting in insufficient stability and interpretability. Furthermore, it is difficult to correct deviations promptly when the model performs poorly in a specific domain. Conversely, relying entirely on expert manual weighting without using any large language model, while providing a clear source of weights, is highly subjective, difficult to scale and update on a large scale, and involves a huge workload.

[0056] This implementation method adopts a "multi-language model scoring + expert weighted fusion" approach, which makes the edge weights have the advantages of high automation and wide coverage, while ensuring the reliability in the medical field through expert weight adjustment and post-correction. It is significantly better than the above-mentioned alternatives in terms of stability, scalability and maintainability, which is one of the core technical features of this invention.

[0057] In one alternative implementation, the weighted membership function is a fuzzy membership function, used to achieve a fuzzy mapping from causal intensity values ​​to causal intensity levels.

[0058] Step S304: Generate a TCM clinical diagnosis and treatment model based on the network topology, weight matrix, and weight membership function.

[0059] The method for constructing a TCM clinical diagnosis and treatment model provided in this embodiment extracts semantic features from a medical knowledge base to construct a network topology. It then uses a multilingual model and expert fusion approach to determine the causal strength values ​​of directed edges, thereby generating a TCM clinical diagnosis and treatment model. This method fully utilizes the understanding capabilities of large language models for massive amounts of text and the reliability of expert experience, transforming unstructured medical guidelines into a computable and traceable standardized knowledge graph. This not only effectively solves the problem of inconsistent concepts in multi-source guidelines but also enhances the interpretability of the model through fuzzy mapping of weight membership functions. This makes the diagnosis and treatment recommendations both reliable and dynamically updated, significantly improving the accuracy, stability, and clinical applicability of the TCM clinical diagnosis and treatment model.

[0060] This embodiment provides a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and a clinical decision support system, which can be used on the aforementioned mobile terminals, such as mobile phones and tablets. Figure 9 This is a first flowchart of a clinical decision support method according to an embodiment of the present invention, such as... Figure 9 As shown, the process includes the following steps: Step S901: Obtain the patient's clinical information; the clinical information includes clinical diagnosis and treatment data, clinical diagnosis and treatment scenarios, and clinical context information.

[0061] Step S902: Map the clinical information to the TCM clinical diagnosis and treatment model constructed using the above method.

[0062] Step S903: Based on the clinical diagnosis and treatment scenario and clinical context information, adjust the node activation state and directed edge weights in the TCM clinical diagnosis and treatment model to suppress nodes that do not match the clinical diagnosis and treatment scenario and clinical context information, and enhance nodes and directed edges that are compatible with the clinical diagnosis and treatment scenario and clinical context information, thereby obtaining the initial state vector and effective inference subgraph.

[0063] In other words, after completing the structural construction and edge weight determination of GM-FCM, a gating and context modulation mechanism is first introduced during the inference phase to constrain and dynamically adjust the inference process, such as... Figure 10 As shown. Specifically, before mapping case information to an initial state vector, the system performs gating control or weight modulation on some nodes and edges in GM-FCM based on the current application scenario and available context information. The context information includes at least one or more of the following: clearly defined or highly reliable syndrome information; known disease diagnosis or syndrome classification information; specific disease-related clinical scenario constraints (such as only performing syndrome identification or only recommending treatment plans); and applicable conditions explicitly defined in the guidelines. During the gating process, the system can suppress nodes that do not conform to the current context (e.g., reduce their initial activation value or temporarily block their input), increase the weight of specific symptoms and syndromes, reduce the weight of common symptoms, or enhance nodes and edges that are highly relevant to the current context (e.g., increase their initial activation value or edge weight), thereby forming a context-aware initial state vector and an effective reasoning subgraph.

[0064] By introducing gating and context modulation mechanisms in the initial stage of inference, we can avoid the indiscriminate propagation of activation values ​​across the entire graph by GM-FCM, reduce interference from irrelevant nodes, make the inference process closer to the actual clinical decision-making situation, and improve the relevance and stability of the inference results.

[0065] While performing inference directly on the complete FCM without any gating or context modulation is simple to implement, it is susceptible to interference from irrelevant nodes, resulting in noisy results and difficulty in reflecting clinical context under multi-task or partially known information conditions. Although manually setting rules to permanently block certain nodes provides clear logic, the rules are rigid, difficult to adapt to different cases and application scenarios, and have poor scalability. This implementation introduces scenario constraints while maintaining inference flexibility through dynamic gating and context modulation.

[0066] Step S904: Based on the initial state vector, the effective reasoning subgraph, and the clinical diagnosis and treatment data, perform iterative reasoning on the TCM clinical diagnosis and treatment model until the state of the TCM clinical diagnosis and treatment model converges, and obtain the reasoning result.

[0067] The clinical decision support method provided in this embodiment dynamically adjusts the model by introducing clinical diagnosis and treatment scenarios and contextual information. This effectively suppresses the interference of irrelevant nodes, enhances the contribution of nodes and directed edges that match the current case, and makes the model more consistent with the actual clinical situation. Iterative reasoning by combining the initial state vector and the effective reasoning subgraph improves the pertinence and accuracy of diagnosis.

[0068] This embodiment provides a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and a clinical decision support system, which can be used on the aforementioned mobile terminals, such as mobile phones and tablets. Figure 11 This is a second flowchart of a clinical decision support method according to an embodiment of the present invention, such as... Figure 11 As shown, the process includes the following steps: Step S1101: Obtain the patient's clinical information; the clinical information includes clinical diagnosis and treatment data, clinical diagnosis and treatment scenarios, and clinical context information.

[0069] Step S1102: Map the clinical information to the TCM clinical diagnosis and treatment model constructed using the above method.

[0070] Step S1103: Based on the clinical diagnosis and treatment scenario and clinical context information, adjust the node activation state and directed edge weights in the TCM clinical diagnosis and treatment model to suppress nodes that do not match the clinical diagnosis and treatment scenario and clinical context information, and enhance nodes and directed edges that are compatible with the clinical diagnosis and treatment scenario and clinical context information, so as to obtain the initial state vector and effective inference subgraph.

[0071] Step S1104: Based on the initial state vector, the effective reasoning subgraph, and the clinical diagnosis and treatment data, perform iterative reasoning on the TCM clinical diagnosis and treatment model until the state of the TCM clinical diagnosis and treatment model converges, and obtain the reasoning result.

[0072] In some optional implementations, the TCM clinical diagnosis and treatment model is iteratively reasoned based on the initial state vector, the effective reasoning subgraph, and the clinical diagnosis and treatment data until the state of the TCM clinical diagnosis and treatment model converges, and the reasoning result is obtained by the following steps S11041 to S11045.

[0073] Step S11041: Using the initial state vector as the initial input for iterative reasoning, in each iteration of the TCM clinical diagnosis and treatment model, the weighted input value of the node in the current round is calculated based on the directed edge weights and node activation states in the effective reasoning subgraph.

[0074] Step S11042: Using the hyperbolic tangent function with zero center as the node activation function, perform mapping calculation on the weighted input values ​​to obtain the intermediate activation value of the node in the current round.

[0075] In other words, a zero-centered hyperbolic tangent function is used as the node activation function, and a damped update strategy is introduced during the state update process. In each iteration, the new activation value of the node is not only calculated from the weighted input, but also merged with the activation state of the previous round according to a preset ratio, thereby suppressing oscillations and enhancing numerical stability. Through the above method, the system can make the state vector converge to a stable state within a finite number of iterations, and the positive and negative causal relationships can be effectively distinguished numerically. Figure 12 It demonstrates the process of case input, status initialization, iterative update, and result output.

[0076] Initially, the standard Logistic-Sigmoid was used as the node activation function, and the state vector S in the t-th round was... (t) =[S1 (t) S2 (t) ,…, S n (t) ] T Updated to (5) in This is the normalized weight matrix. Iteration proceeds when... (6) It may terminate when the maximum number of steps t=100 is reached.

[0077] Experimental results show that regardless of the initial vector setting, the system converges to the same fixed point around the 32nd iteration, indicating that the sigmoid function is prone to early saturation and loss of input diversity in the FCM network structure of this disease. To enhance the model's ability to preserve initial activation differences, a scaled Hyperbolic-Tangent activation function with a zero center and a gentler slope near the origin was used instead. (7) Meanwhile, to avoid completely discarding historical information in a single iteration, a damping coefficient is added. The updated formula is corrected to (8) In this way, the limitations of the initial Sigmoid activation scheme are resolved. The node state can respond to new inputs and retain some values ​​from the previous round, ensuring that information from different initial conditions can be slowly accumulated and amplified in multiple iterations.

[0078] The new iteration termination condition has been adjusted to (9) Or reach the maximum number of iterations T max =300.

[0079] Under this setup, the disease-specific FCM model exhibits significant differentiation with different initial activation vectors and typically converges within 150–250 rounds. Compared to the Sigmoid scheme, it preserves input diversity without excessively increasing computation time. After parameter tuning, the final values ​​were determined to be β=0.5, λ=0.4, Tmax=300, and ε=10. The iterative strategy of step 4 balances convergence speed and input sensitivity, providing a reliable methodological foundation for subsequent inference result analysis. The equilibrium vector C18 after convergence is the inference result.

[0080] This improved inference mechanism solves the problems of numerical oscillation, saturation, or slow convergence that traditional FCM is prone to in complex networks, enabling the model to maintain stable and efficient inference performance in GM-FCM containing multiple types of nodes and multiple causal paths.

[0081] While traditional sigmoid or linear activation functions are simple to implement, they have weak expressive power for positive and negative relationships, are prone to saturation, and have insufficient discriminative power in inference. If the standard FCM update formula is used without introducing a damping term, it is prone to oscillation in large-scale graph structures, making stable convergence difficult to guarantee.

[0082] The embodiments of the present invention, by combining the zero-center activation function with damped updates, outperform the above-mentioned alternatives in terms of numerical stability, convergence speed, and discriminative ability, which is one of the important technical improvements of the present invention.

[0083] Step S11043: Obtain the final activation value of the node from the previous round.

[0084] Step S11044: The intermediate activation value of the node in the current round and the final activation value of the node in the previous round are weighted and fused according to the preset damping coefficient to obtain the final activation value of the node in the current round.

[0085] Step S11045: Repeat the above iterative process until the TCM clinical diagnosis and treatment model converges and the inference result is obtained.

[0086] In one optional implementation, after obtaining the reasoning result, the method further includes: dividing the nodes in the TCM clinical diagnosis and treatment model into multiple functional node groups according to their functional attributes, wherein each functional node group includes at least a symptom node group, a syndrome node group, and a treatment node group; for each functional node group, performing an intra-group competitive decision-making operation: calculating the sum of the activation values ​​of all nodes in the functional node group to obtain a normalized benchmark; based on the normalized benchmark, normalizing the activation values ​​of each node in the functional node group to obtain the relative importance of each node; selecting nodes with a relative importance higher than a preset threshold as candidate recommended nodes, and sorting the candidate recommended nodes according to their relative importance to obtain the diagnosis and treatment recommendation results for the functional node group.

[0087] In other words, after the GM-FCM inference converges, this embodiment of the invention introduces an intra-group competition mechanism to achieve multi-task decision output based on a unified inference result. Specifically, the system divides the nodes in GM-FCM into several functional groups according to node type, including disease node groups, syndrome node groups, treatment principle node groups, prescription node groups, and traditional Chinese medicine node groups. For each node group, the system normalizes its activation values ​​and uses softmax, threshold filtering, or other competitive strategies to create a relative competitive relationship between nodes within the group, thereby outputting recommendation results with clear ranking and high discrimination. In this way, the system can simultaneously output decision results at multiple levels during the same inference process, including disease diagnosis recommendations, syndrome identification results, and recommended lists of treatment principles, prescriptions, and traditional Chinese medicines.

[0088] By avoiding meaningless comparisons between different types of nodes through intra-group competition, the output of multiple tasks can maintain internal consistency and interpretability under a unified knowledge framework, effectively supporting the integrated decision-making needs in complex disease diagnosis and treatment scenarios.

[0089] If node activation values ​​are used directly for global sorting, the results will be mixed due to the different dimensions and semantics of different node types, leading to poor clinical usability. Another approach is to build a separate model for each task category. While this single-model approach is logically clear, it suffers from model redundancy, high maintenance costs, and potentially contradictory results across different tasks. The implementation of this invention achieves multi-task output within a single GM-FCM framework through intra-group competition, offering significant advantages in consistency, maintainability, and engineering efficiency.

[0090] Although traditional FCM models have a certain degree of interpretability in theory, in practice they often only output the final activation value or recommended ranking to doctors, lacking a clear display of key nodes, key paths and their correspondence with guideline clauses: doctors find it difficult to intuitively understand which symptoms, syndromes and causal relationships drive a certain recommendation result; the reasoning trajectories under different cases are not easy to compare, which is not conducive to multidisciplinary discussions and teaching and training; and it is impossible to clearly explain why the system makes such a recommendation in the natural language that doctors are used to.

[0091] In one optional implementation, after sorting the candidate recommendation nodes according to their relative importance to obtain the diagnosis and treatment recommendation results of the functional node group, the method further includes: selecting high-activation nodes and high-weight edges related to the diagnosis and treatment recommendation results from the convergence state and weight matrix of the TCM clinical diagnosis and treatment model; constructing representative diagnosis and treatment paths based on the high-activation nodes and high-weight edges, generating structured path data corresponding to the representative diagnosis and treatment paths; and generating natural language interpretations based on the structured path data.

[0092] In other words, from the convergent state C18 and the weight matrix W, highly activated nodes and high-weight edges relevant to the final recommendation result are selected and input into the large language model to construct one or more representative diagnostic and treatment paths, such as "symptoms / examination → syndrome → disease → treatment principle → prescription → traditional Chinese medicine," forming structured path data. Then, this structured path data and relevant guideline clause identifiers are used as input to call the large language model to generate natural language explanations, explaining: the reasons for recommending a certain diagnosis or syndrome; the significant contribution of key symptoms, syndromes, and high-weight edges to the recommendation result; and the consistency between the recommended scheme and relevant guidelines. Finally, the system presents the recommendation result, local FCM subgraphs, and natural language explanations to the user in a combined text and image format.

[0093] The significance of this step lies in providing doctors with traceable and understandable decision-making basis, thereby improving the acceptability and trustworthiness of the system in disease and traditional Chinese medicine scenarios.

[0094] While simply outputting numerical scores and lists without providing path and natural language explanations is simple to implement, it lacks interpretability and hinders clinical application. Displaying FCM subgraphs without generating textual descriptions using a large language model offers good visualization, but physicians need to understand the graph structure themselves, posing a high barrier to entry for non-graph model professionals. This invention's implementation combines "graph structure path + large language model explanation," simultaneously considering structural visualization and natural language expression, thus surpassing the aforementioned solutions in both interpretability and ease of use.

[0095] Based on FCM reasoning, this invention proposes a new reasoning mechanism, GM-FCM. By combining a large language model to extract structured data from highly activated nodes and high-weight edges and generate natural language, it can output explanatory text and graphical paths corresponding to the guide content, significantly improving the interpretability and acceptability of the system in disease and TCM diagnosis and treatment scenarios.

[0096] In summary, this invention provides a method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and a clinical decision support system, mainly implemented by executing a computer program on a server or other computing device. The TCM clinical diagnosis and treatment model includes, for example... Figure 13 As shown, the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and the clinical decision support system are as follows: Figure 14 As shown.

[0097] This invention constructs clinical practice guidelines as a fuzzy cognitive graph and introduces a multi-task output mechanism involving gating and context modulation, improved activation functions and damped updates, and intra-group competition during the reasoning process. This enables unified modeling and interpretable reasoning for multi-level decision-making tasks in the disease diagnosis and treatment process.

[0098] From the perspective of reasoning mechanism, gating and context modulation mechanisms enable the reasoning process to dynamically constrain the propagation path based on known syndromes, diseases, or application scenarios, reducing interference from irrelevant nodes; improved activation functions and damped update mechanisms enhance the numerical stability and discriminative ability of complex causal graphs under multiple iterations; and intra-group competition mechanisms achieve consistent outputs for multiple tasks such as diagnosis, syndromes, treatment methods, prescriptions, and traditional Chinese medicine based on unified reasoning results, avoiding semantic confusion caused by cross-type sorting.

[0099] To verify the effectiveness of this invention, task-level evaluations were conducted on the TCM clinical diagnosis and treatment model and various comparative methods under multi-task settings T1–T5, such as… Figure 15 As shown, the comparison methods include similarity-based kNN, linear model EN-LogReg, rule-based methods, and large language model-only (LLM-only) and retrieval enhancement (RAG-CDS).

[0100] Multi-task experimental results based on real disease cases and guidelines demonstrate that the proposed solution exhibits significant advantages in key decision-making stages centered on syndrome differentiation. For example, in the "symptom → syndrome" task, it achieves an MRR of 0.74 and a Hit@5 of 0.99; in the "syndrome → treatment" task, it achieves an MRR of 0.67 and a Hit@5 of 0.93; and in the "syndrome → traditional Chinese medicine" task, it achieves an MRR of 0.81 and a Hit@1 of 0.71. Overall, it outperforms linear models, similarity methods, and comparison schemes that rely solely on large language models or retrieval enhancement.

[0101] The above experimental results verify the effectiveness, stability, and interpretability of the technical solution of this invention in multi-level disease decision support from the perspective of efficacy. They demonstrate that through the synergistic design of structured causal reasoning and context modulation mechanisms, it is possible to achieve comprehensive performance superior to existing technologies in complex TCM disease diagnosis and treatment scenarios, and it has high practical application value and promotion prospects.

[0102] In summary, the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and the clinical decision support system provided by the embodiments of the present invention have the following beneficial effects: (1) It realizes the systematic transformation of knowledge from multiple TCM clinical practice guidelines for diseases into large-scale directed fuzzy cognitive graphs, enabling the guideline knowledge to have a calculable and reasonable structural representation.

[0103] (2) Multi-level GM-FCM unified knowledge modeling for “symptoms-syndromes-diseases-treatments-prescriptions-traditional Chinese medicines”. The multi-level concepts of TCM disease diagnosis and treatment links are incorporated into the same fuzzy cognitive graph framework, explicitly expressing cross-level causal relationships and uncertainties, and realizing the unified carrying of multi-hop reasoning from symptoms to syndromes, and from syndromes to treatments / prescriptions.

[0104] (3) Weight determination mechanism of multi-language model + expert evaluation. To address the problem of the difficulty in uniformly quantifying the weights of FCM, structured prompt words are used to score the relation strength of multiple large language models, and the weighted fusion and normalization are combined with the expert evaluation of the model reliability to obtain a more stable, reusable and maintainable weight matrix, thereby reducing the subjective bias and cost caused by pure manual weighting.

[0105] (4) Gating and context-modulated reasoning mechanism. Context-aware gating is introduced in the reasoning initialization stage. Effective reasoning subgraphs are dynamically selected and irrelevant path propagation is suppressed based on known syndrome / disease information or specific task scenarios, making the reasoning closer to the clinical context and improving the relevance and stability of the results.

[0106] (5) Improve the iterative inference mechanism of activation function and damped update. A zero-center activation function is adopted and a damped update strategy is introduced to suppress the oscillation and saturation problems of traditional FCM, improve the numerical stability, convergence efficiency and discriminative ability of complex causal graph inference, and adapt to multi-task and multi-level graph structures.

[0107] (6) Intra-group competition and multi-task consistent output mechanism. Diseases, syndromes, treatments, prescriptions and Chinese medicines are grouped by type. Intra-group normalization and competitive ranking are implemented at the convergence state to avoid meaningless comparisons across types, realize multi-task collaborative output under a single reasoning framework, and ensure the internal consistency of decision results at different levels.

[0108] (7) Structured interpretation based on critical path extraction and natural language generation based on large language model. The critical path of diagnosis and treatment composed of high activation nodes and high weight edges is extracted from the converged GM-FCM, and the large language model is called to generate natural language interpretation consistent with the reasoning path, so as to realize the unified output of "traceable causal evidence + readable clinical interpretation" and improve the credibility and usability of the system.

[0109] (8) Modular system implementation for research and clinical integration. Knowledge acquisition, mapping, weighting, reasoning, multi-task output and interpretation generation are modularly encapsulated, which facilitates verification in research evaluation environments and also facilitates deployment with clinical systems such as HIS / EMR, and has good engineering feasibility and scalability.

[0110] This embodiment also provides a device for constructing a TCM clinical diagnosis and treatment model. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0111] This embodiment provides a device for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines, such as... Figure 16 As shown, it includes a feature extraction module 1601, a weight matrix construction module 1602, a level mapping module 1603, and a processing module 1604.

[0112] The feature extraction module 1601 is used to extract semantic features representing disease symptoms, syndromes and treatment plans from the medical knowledge base, and to construct a network topology structure containing nodes and directed edges between nodes, wherein the nodes include symptom nodes, syndrome nodes and treatment nodes.

[0113] The weight matrix construction module 1602 is used to determine the causal strength values ​​of directed edges in the network topology using a preset language model, and obtain the weight matrix. The rank mapping module 1603 is used to define the weight membership function of directed edges to map the causal strength values ​​of directed edges to causal strength ranks.

[0114] Processing module 1604 is used to generate a TCM clinical diagnosis and treatment model based on network topology, weight matrix and weight membership function.

[0115] In some optional implementations, the weight matrix construction module is specifically used to: evaluate the causal relationship strength of the same directed edge using multiple language models respectively, and obtain multiple initial evaluation weights for each directed edge; obtain the credibility score of each language model, and determine the weight coefficient of each language model based on the credibility score of each language model; and perform weighted fusion of the multiple initial evaluation weights belonging to the same directed edge based on the weight coefficient of each language model to obtain the causal strength value of each directed edge.

[0116] In some optional implementations, the weighted membership function is a fuzzy membership function, used to achieve a fuzzy mapping from causal strength values ​​to causal strength levels.

[0117] This embodiment provides a clinical decision support device, such as Figure 17 As shown, it includes an acquisition module 1701, an input module 1702, an adjustment module 1703, and a reasoning module 1704.

[0118] The acquisition module 1701 is used to acquire the clinical information of patients with diseases; the clinical information includes clinical diagnosis and treatment data, clinical diagnosis and treatment scenarios and clinical context information.

[0119] Input module 1702 is used to map clinical information to the TCM clinical diagnosis and treatment model constructed using the above method.

[0120] The adjustment module 1703 is used to adjust the activation state of nodes and the weight of directed edges in the TCM clinical diagnosis and treatment model according to the clinical diagnosis and treatment scenario and clinical context information, so as to suppress nodes that do not match the clinical diagnosis and treatment scenario and clinical context information, enhance nodes and directed edges that are compatible with the clinical diagnosis and treatment scenario and clinical context information, and obtain the initial state vector and effective inference subgraph.

[0121] The reasoning module 1704 is used to iteratively reason about the TCM clinical diagnosis and treatment model based on the initial state vector, the effective reasoning subgraph, and clinical diagnosis and treatment data until the state of the TCM clinical diagnosis and treatment model converges and the reasoning result is obtained.

[0122] In some optional implementations, the inference module 1704 is specifically used to: use the initial state vector as the initial input for iterative inference; in each iteration of the TCM clinical diagnosis and treatment model, calculate the weighted input value of the node in the current round based on the directed edge weights and node activation states in the effective inference subgraph; use the hyperbolic tangent function with zero center as the node activation function to map and calculate the weighted input value to obtain the intermediate activation value of the node in the current round; obtain the final activation value of the node in the previous round; weight and fuse the intermediate activation value of the node in the current round with the final activation value of the node in the previous round according to a preset damping coefficient to obtain the final activation value of the node in the current round; and repeatedly execute the above iterative process until the TCM clinical diagnosis and treatment model converges to obtain the inference result.

[0123] In some optional implementations, the TCM clinical diagnosis and treatment model and clinical auxiliary decision-making system device based on clinical practice guidelines also includes a post-processing module. After obtaining the inference results, the post-processing module is used to: divide the nodes in the TCM clinical diagnosis and treatment model into multiple functional node groups according to their functional attributes. The functional node groups include at least symptom node groups, syndrome node groups, and treatment node groups; for each functional node group, perform intra-group competitive decision-making operations: calculate the sum of the activation values ​​of all nodes in the functional node group to obtain a normalized benchmark; based on the normalized benchmark, normalize the activation values ​​of each node in the functional node group to obtain the relative importance of each node; select nodes with relative importance higher than a preset threshold as candidate recommended nodes, and sort the candidate recommended nodes according to their relative importance to obtain the diagnosis and treatment recommendation results for the functional node group.

[0124] In some optional implementations, after sorting the candidate recommendation nodes according to their relative importance to obtain the diagnosis and treatment recommendation results of the functional node group, the post-processing module is used to: select high-activation nodes and high-weight edges related to the diagnosis and treatment recommendation results in the convergence state and weight matrix of the TCM clinical diagnosis and treatment model; construct representative diagnosis and treatment paths based on the high-activation nodes and high-weight edges, and generate structured path data corresponding to the representative diagnosis and treatment paths; and generate natural language interpretations based on the structured path data.

[0125] The apparatus for constructing a TCM clinical diagnosis and treatment model and / or the apparatus for a clinical auxiliary decision-making system based on a TCM clinical diagnosis and treatment model provided in this embodiment of the invention can execute the method for constructing a TCM clinical diagnosis and treatment model and / or the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines and the clinical auxiliary decision-making system provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0126] Figure 18 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0127] The following is a detailed reference. Figure 18 This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 1801, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 1802 or a program loaded from memory 1808 into random access memory (RAM) 1803. The RAM 1803 also stores various programs and data required for the operation of the electronic device. The processor 1801, ROM 1802, and RAM 1803 are interconnected via a bus 1804. An input / output (I / O) interface 1805 is also connected to the bus 1804.

[0128] Typically, the following devices can be connected to the I / O interface 1805: input devices 1806 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 1807 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; memory devices 1808 including, for example, magnetic tape, hard disk, etc.; and communication devices 1809. Communication devices 1809 allow electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 18 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0129] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 1809, or installed from a memory 1808, or installed from a ROM 1802. When the computer program is executed by the processor 1801, it performs the functions defined in the method for constructing a TCM clinical diagnosis and treatment model and the clinical auxiliary decision-making system of the embodiments of the present invention.

[0130] Figure 18 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0131] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the method for constructing a traditional Chinese medicine clinical diagnosis and treatment model and the clinical auxiliary decision-making system shown in the above embodiments are implemented.

[0132] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0133] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines, characterized in that, The method includes: Extract semantic features representing disease symptoms, syndromes and treatment plans from a medical knowledge base, and construct a network topology containing nodes and directed edges between nodes, wherein the nodes include symptom nodes, syndrome nodes and treatment nodes. The causal strength values ​​of the directed edges in the network topology are determined using a preset language model, and a weight matrix is ​​obtained. Define the weighted membership function of the directed edge to map the causal strength value of the directed edge to the causal strength level; Based on the network topology, the weight matrix, and the weight membership function, a TCM clinical diagnosis and treatment model is generated.

2. The method according to claim 1, characterized in that, The step of determining the causal strength values ​​of directed edges in the network topology using a preset language model to obtain a weight matrix includes: The causal relationship strength of the same directed edge is evaluated using multiple language models to obtain multiple initial evaluation weights for each directed edge. Obtain the credibility score for each of the language models, and determine the weight coefficient for each of the language models based on the credibility score. The weight coefficients of each language model are used to weight and fuse the multiple initial evaluation weights belonging to the same directed edge to obtain the causal strength value of each directed edge.

3. The method according to claim 1, characterized in that, The weighted membership function is a fuzzy membership function, used to realize the fuzzy mapping from the causal intensity value to the causal intensity level.

4. A clinical decision support method, characterized in that, The method includes: Obtain clinical information of patients with diseases; wherein the clinical information includes clinical diagnosis and treatment data, clinical diagnosis and treatment scenarios, and clinical context information; The clinical information is mapped to a traditional Chinese medicine clinical diagnosis and treatment model constructed using any one of claims 1 to 3; Based on the clinical diagnosis and treatment scenario and clinical context information, the activation state of nodes and the weight of directed edges in the TCM clinical diagnosis and treatment model are adjusted to suppress nodes that do not match the clinical diagnosis and treatment scenario and clinical context information, and to enhance nodes and directed edges that are compatible with the clinical diagnosis and treatment scenario and clinical context information, thereby obtaining an initial state vector and an effective inference subgraph. Based on the initial state vector, the effective reasoning subgraph, and the clinical diagnosis and treatment data, the TCM clinical diagnosis and treatment model is iteratively reasoned until the state of the TCM clinical diagnosis and treatment model converges, and the reasoning result is obtained.

5. The method according to claim 4, characterized in that, The iterative reasoning process based on the initial state vector, the effective reasoning subgraph, and the clinical diagnosis and treatment data, until the state of the TCM clinical diagnosis and treatment model converges, yields the following reasoning results: Using the initial state vector as the initial input for iterative reasoning, in each iteration of the TCM clinical diagnosis and treatment model, the weighted input value of the node in the current round is calculated based on the directed edge weights and node activation states in the effective reasoning subgraph. Using the zero-centered hyperbolic tangent function as the node activation function, the weighted input values ​​are mapped and calculated to obtain the node intermediate activation values ​​for the current round; Get the final activation value of the node from the previous round; The intermediate activation value of the node in the current round is weighted and fused with the final activation value of the node in the previous round according to a preset damping coefficient to obtain the final activation value of the node in the current round. The above iterative process is repeated until the TCM clinical diagnosis and treatment model converges, and the inference result is obtained.

6. The method according to claim 4, characterized in that, After obtaining the reasoning result, it also includes: The nodes in the TCM clinical diagnosis and treatment model are divided into multiple functional node groups according to their functional attributes. The functional node groups include at least symptom node groups, syndrome node groups, and treatment node groups. For each functional node group, an intra-group competitive decision-making operation is performed: the sum of the activation values ​​of all nodes in the functional node group is calculated to obtain a normalized benchmark; based on the normalized benchmark, the activation values ​​of each node in the functional node group are normalized to obtain the relative importance of each node; nodes with relative importance higher than a preset threshold are selected as candidate recommended nodes, and the candidate recommended nodes are sorted according to the relative importance to obtain the diagnosis and treatment recommendation results of the functional node group.

7. The method according to claim 6, characterized in that, After sorting the candidate recommendation nodes according to their relative importance to obtain the diagnosis and treatment recommendation results for the functional node group, the method further includes: In the convergence state and weight matrix of the TCM clinical diagnosis and treatment model, select high-activation nodes and high-weight edges that are related to the diagnosis and treatment recommendation results; Based on the highly activated nodes and the highly weighted edges, a representative treatment path is constructed, and structured path data corresponding to the representative treatment path is generated. Generate natural language explanations based on the structured path data.

8. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected and communicate with each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines as described in any one of claims 1 to 3 and / or the clinical auxiliary decision-making method as described in any one of claims 4 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines as described in any one of claims 1 to 3 and / or the clinical decision support method as described in any one of claims 4 to 7.

10. A computer program product, characterized in that, Includes computer instructions for causing a computer to execute the method for constructing a TCM clinical diagnosis and treatment model based on clinical practice guidelines as described in any one of claims 1 to 3 and / or the clinical decision support method as described in any one of claims 4 to 7.

Images