Traditional Chinese medicine famous prescription and famous operation multi-agent collaborative consultation method and system

Abstract

The invention discloses a traditional Chinese medicine famous prescription and operation multi-agent collaborative consultation method and system. The system comprises a medical case knowledge base module, an expert knowledge base module, a medicine knowledge base module, a pre-differentiation module, a multi-agent diagnosis module and the like. The method comprises the following steps: constructing structured experience medical case data, an expert experience knowledge graph and a traditional Chinese medicine knowledge graph knowledge base, carrying out mixed retrieval to carry out pre-differentiation, calling multi-agent consultation according to preliminary disease differentiation and syndrome types, and obtaining an expert diagnosis scheme for risk verification and prescription adjustment optimization. According to the invention, a multi-agent collaborative diagnosis technology is utilized, and a multi-agent architecture for simulating real traditional Chinese medicine consultation is constructed. A plurality of expert agents perform independent diagnosis and complement each other, so that the limitation of a single agent is avoided, and comprehensive and accurate diagnosis is provided. And in combination with expert experience knowledge graph and big language model reasoning, the system can flexibly differentiate symptoms according to the symptoms of the patient, accurately judge disease types and syndrome types, and improve the differentiation accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traditional Chinese medicine collaborative consultation, in particular to a multi-agent collaborative consultation method and system for traditional Chinese medicine prescriptions and techniques. BACKGROUND

[0002] With the landing of large models in various fields, training a traditional Chinese medicine large model using traditional Chinese medicine knowledge and expert medical records is a feasible technical solution. However, the demand for large models in the field of traditional Chinese medicine is not only for prescribing and treating diseases, but also for inheriting and developing the valuable experience and knowledge of famous old traditional Chinese medicine experts. This requires the traditional Chinese medicine large model to accurately identify and distinguish the treatment characteristics of traditional Chinese medicine experts and medical school factions, and to master the differentiation characteristics of different experts. The current traditional Chinese medicine diagnosis and treatment model has three key limitations in this scenario:

[0003] First, the limitation of single-agent decision-making. Existing systems generally use a single model for syndrome differentiation, which inevitably leads to the templateization of treatment plans and the inability to simulate the clinical scenario of "multi-specialist collaborative decision-making" in real consultations. It is difficult to combine the strengths of various treatment plans and summarize experience through comparison. Second, weak knowledge integration capability. Although some newer systems have introduced knowledge graphs, the experience of experts from different factions still exists in isolation in the form of static rule bases, lacking dynamic interaction and collaboration represented by agents. Third, the consultation process is broken. Existing technologies cannot achieve a complete consultation closed loop of "initial diagnosis-expert debate-consensus formation", especially when experts have different opinions. The system often mechanically uses the voting average method, ignoring the theoretical differences between different factions of traditional Chinese medicine.

[0004] For example, the invention application with application number 202010416066.4 discloses a traditional Chinese medicine remote consultation system based on artificial intelligence. The application uses artificial intelligence to realize the remote transmission and standardization of diagnosis of four diagnostic data of observation, listening, questioning and palpation, and can help doctors and patients to accurately diagnose remotely. However, the scheme has the following problems: it uses a single algorithm for syndrome differentiation, which cannot simulate multi-specialist collaborative decision-making, leading to templateization of diagnosis and treatment plans; the knowledge graph is a static rule base, and the experience of different factions is isolated, lacking a dynamic interaction and fusion mechanism; the consultation process is incomplete, and when opinions differ, only the voting average method is used, ignoring the theoretical differences between different factions, making it difficult to form a consensus, unable to inherit the experience and characteristics of famous old traditional Chinese medicine experts, and not breaking through the limitations of existing traditional Chinese medicine diagnosis and treatment models.

[0005] Therefore, there is a need for a multi-agent collaborative consultation system for traditional Chinese medicine prescriptions and techniques that can overcome the above limitations. By introducing a multi-agent collaborative decision-making mechanism, the multi-specialist collaborative process in real consultations is simulated, and the diversification and individualization of treatment plans are achieved. SUMMARY

[0006] In view of the above problems, the present application aims to provide a traditional Chinese medicine prescription and technique multi-agent collaborative consultation method and system, which realizes multi-dimensional and multi-level consultation and collaborative decision-making in the traditional Chinese medicine diagnosis process by integrating multi-agent technology, and provides more comprehensive and accurate diagnosis results.

[0007] The present application provides a traditional Chinese medicine prescription and technique multi-agent collaborative consultation method and system.

[0008] The first aspect is a traditional Chinese medicine prescription and technique multi-agent collaborative consultation system, comprising:

[0009] A medical record knowledge base module is used to collect and store diagnosis and treatment medical records, and construct a structured experience medical record data knowledge base.

[0010] An expert knowledge base module is used to collect and store known expert syndrome differentiation logic and treatment method schemes, and construct a structured expert experience knowledge graph knowledge base.

[0011] A medical knowledge base module is used to extract and store key attributes of traditional Chinese medicine prescriptions, and construct a structured traditional Chinese medicine knowledge graph knowledge base.

[0012] A preliminary syndrome differentiation module is used to perform preliminary disease differentiation and syndrome type judgment according to consultation requirements by using expert experience knowledge graph hybrid retrieval.

[0013] A multi-agent diagnosis module is configured with multiple agents, and performs collaborative consultation according to the preliminary disease differentiation and syndrome type judgment.

[0014] Optionally, the multiple agents comprise:

[0015] A supervising agent is used to coordinate and assign consultation tasks to multiple expert agents, and summarize the diagnosis schemes of the multiple expert agents.

[0016] Multiple expert agents are used to perform individualized diagnosis by assigning consultation tasks and performing weighted retrieval of experience medical record data.

[0017] A review agent is used to perform traditional Chinese medicine knowledge graph risk verification and prescription optimization of the diagnosis scheme.

[0018] The second aspect is a traditional Chinese medicine prescription and technique multi-agent collaborative consultation method, comprising:

[0019] S1, constructing a structured experience medical record data, expert experience knowledge graph and traditional Chinese medicine knowledge graph knowledge base;

[0020] S2, performing preliminary disease differentiation and syndrome type judgment by using expert experience knowledge graph hybrid retrieval of the knowledge base according to consultation requirements.

[0021] S3, according to the preliminary diagnosis and syndrome type judgment of consultation, call multiple multi-agent for collaborative consultation, get the expert diagnosis scheme of risk check and prescription optimization.

[0022] Optionally, the pre-diagnosis in S2 includes the steps of:

[0023] S21, according to the consultation disease, using mixed retrieval strategy, retrieving relevant syndrome experience from expert experience knowledge graph as large language model prompt text;

[0024] S22, based on the optimized large language model, output the syndrome expression conforming to the "standard of diagnosis and efficacy of traditional Chinese medicine disease and syndrome".

[0025] Optionally, the expert experience knowledge graph includes:

[0026] Using a large language model to summarize expert diagnosis and comments from tens of thousands of real medical records, combining the Graph RAG framework to extract entity relationships, and generating community summaries to write into the knowledge graph.

[0027] Optionally, the multiple multi-agent collaborative consultation in S3 includes the steps of:

[0028] S31, call the supervision agent, and distribute the consultation task to multiple expert agents;

[0029] S32, each expert agent performs individualized diagnosis, and the results are summarized to the supervision agent to complete a round of diagnosis;

[0030] S33, when the predetermined number of diagnosis rounds is reached or the expert agent has no more questions, the supervision agent summarizes the expert diagnosis scheme and delivers it to the review agent;

[0031] S34, the review agent performs risk check and prescription optimization on the diagnosis scheme based on the traditional Chinese medicine knowledge graph.

[0032] Optionally, in S32, when the expert agent receives the diagnosis task assigned by the supervision agent:

[0033] Relying on different expert agents to mount special knowledge bases and weighted retrieval of experience medical record data as large language model RAG prompt text, and encouraging the large language model to refer to experience medical record data during diagnosis.

[0034] Optionally, the risk check and prescription optimization of the review agent in S34 includes:

[0035] Based on the traditional Chinese medicine knowledge graph, a traditional Chinese medicine contraindication graph is constructed, wherein the edge weight between the contraindication nodes represents the strength of the medicinal material contraindication, and the medicinal material set in the diagnosis scheme is searched to obtain the contraindicated drugs, and the contraindicated drugs are replaced to optimize the diagnosis scheme.

[0036] Third aspect: An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method provided in the second aspect when executing the program.

[0037] Fourth aspect: A non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the steps of the method provided in the second aspect.

[0038] Advantages of the present application:

[0039] 1. The present application adopts multi-agent collaborative diagnosis technology, simulates the real TCM consultation scene by constructing a multi-agent architecture. In this architecture, multiple expert agents independently diagnose and complement each other, effectively avoiding the limitations of single-agent decision-making, thereby providing more comprehensive and accurate diagnosis results. The flexible syndrome differentiation method based on expert experience, combined with the reasoning ability of expert experience knowledge graph and large language model, can flexibly adjust the differentiation direction according to the patient's specific symptoms and signs, accurately judge the disease type and syndrome type. When the differentiation elements are insufficient, the system will actively ask the user questions to improve the medical record information and further improve the accuracy of differentiation.

[0040] 2. The present application deeply applies knowledge graph technology and constructs a multi-dimensional knowledge base including TCM knowledge graph, expert experience knowledge graph, etc. These knowledge bases cover rich content such as TCM theory, famous doctors, famous prescriptions, and expert differentiation logic. These knowledge graphs not only provide a solid knowledge base for intelligent agents, but also realize the deep integration and inheritance of knowledge through dynamic interaction and collaboration. Converting expert experience into structured data in the knowledge graph not only preserves the valuable experience of famous old TCM doctors, but also enables it to be dynamically called and inherited in the system, thereby promoting the dissemination and development of TCM knowledge.

[0041] 3. The present application realizes full-process closed-loop management, and the system can complete the complete consultation closed loop from initial diagnosis, expert argumentation to consensus formation. During the diagnosis process, the supervision agent is responsible for coordinating task allocation, summarizing expert opinions, interacting with patients to supplement information, and other key links to ensure the coherence and effectiveness of the consultation process. Even when there are differences in expert opinions, the system does not simply use the voting mean method, but through multiple rounds of interaction and expert argumentation, a more targeted and scientific diagnosis conclusion is formed. After the consultation, the risk review agent strictly evaluates and optimizes the treatment plan. Through the construction of the contraindication relationship graph and multi-dimensional conflict detection, potential contraindication risks are discovered in a timely manner, and intelligent prescription optimization suggestions are provided. This link not only ensures the safety of patients, but also further improves the quality of treatment plans.

[0042] 4, The application adopts an efficient intelligent agent cooperation mechanism, a system adopts a multi-agent architecture, a supervisory intelligent agent can quickly select a suitable expert to form an expert group according to a disease type and a complaint, and reasonably allocate a diagnosis task. In an independent diagnosis process, the expert intelligent agent can quickly search a historical medical record most similar to the current patient as a reference, and improve diagnosis efficiency. In the diagnosis process, the system dynamically interacts with the patient, supplements medical record information through the questioning mode, and ensures the accuracy and integrity of the diagnosis. Meanwhile, in the risk review stage, the system generates a structured inquiry template for high-risk unannounced contraindications, and interacts with the patient to improve user experience. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 Fig. 1 is a structural schematic diagram of a collaborative consultation system of the application;

[0044] Figure 2 Fig. 2 is a flow schematic diagram of a collaborative consultation method of the application;

[0045] Figure 3 Fig. 3 is an application schematic diagram of the collaborative consultation method in the embodiment of the application;

[0046] Figure 4 Fig. 4 is a schematic diagram of an expert experience knowledge graph in the embodiment of the application;

[0047] Figure 5 Fig. 5 is an application schematic diagram of a consultation system in the embodiment of the application;

[0048] Figure 6 Fig. 6 is a structural schematic diagram of an electronic device of the application. DETAILED DESCRIPTION

[0049] The embodiments of the application are described in detail below, examples of which are shown in the drawings, wherein the same or similar symbols represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the application, and cannot be understood as a limitation of the application.

[0050] The current TCM collaborative diagnosis and treatment faces the problems that a single decision model limits the diversity of treatment schemes, cannot simulate the multi-expert collaborative decision-making in actual consultation, the knowledge fusion ability is insufficient, lacks dynamic interaction, the consultation process is not coherent, and the closed loop from initial diagnosis to consensus cannot be completely realized.

[0051] In view of the above problems, the application provides a multi-agent collaborative consultation system of TCM famous prescriptions and famous techniques, Figure 1 Fig. 1 is a structural schematic diagram of a system provided by the embodiment of the application, and the system comprises a medical record knowledge base module, an expert knowledge base module, a medical knowledge base module, a pre-diagnosis module and a multi-agent diagnosis module, etc., wherein:

[0052] The medical record knowledge base module is used to collect and store medical records, and to construct a structured experience medical record data knowledge base.

[0053] In this medical record knowledge base module, not only the multi-vector data of medical records are stored, but also the structured medical record records are embedded attribute by attribute. In this way, each medical record can be transformed into a multi-dimensional embedding matrix through joint representation;

[0054] The medical record knowledge base uses multi-vector embedding representation, which retains more disease details and has higher query accuracy than the overall embedding of medical records.

[0055] For example: According to different diseases, four diagnostic information embedding representation is adopted, the four diagnostic information is extracted into specific items, four diagnostic objects are formed, each attribute of the object is independently embedded to form a four diagnostic embedding object, and a vector database is used for storage. In the query process, first, the query medical record is embedded with four diagnostic information, the similarity distance of each four diagnostic item is calculated, and then the weighted sum of the calculation results is calculated according to the weight matrix, and the top k ranking is extracted as the query result.

[0056] The expert knowledge base module is used to collect and store the syndrome differentiation logic and treatment plan of famous experts, and to construct a structured expert experience knowledge graph knowledge base;

[0057] The expert experience knowledge graph is mainly used to store the syndrome differentiation logic and treatment plan from famous experts. It is not only a collection of information, but also a structured knowledge base that reflects the wisdom and clinical experience of experts, so that these valuable knowledge can be effectively managed and utilized. In this way, a structured expert experience knowledge graph knowledge base can be constructed, so that these valuable knowledge can be systematically organized and stored.

[0058] The medical knowledge base module is used to extract and store the key attributes of traditional Chinese medicine prescriptions, and to construct a structured traditional Chinese medicine knowledge graph knowledge base. It is mainly responsible for extracting and storing the key attribute information of traditional Chinese medicine prescriptions. Through this process, a structured traditional Chinese medicine knowledge graph knowledge base can be constructed. This knowledge base not only contains rich information of traditional Chinese medicine prescriptions, but also helps researchers and medical workers quickly retrieve and utilize these valuable knowledge resources.

[0059] The traditional Chinese medicine knowledge graph is constructed on the basis of the traditional Chinese medicine encyclopedia by deeply analyzing and extracting the key attributes of traditional Chinese medicine prescriptions. This knowledge graph not only covers the compatibility relationship of traditional Chinese medicine prescriptions, but also records important information such as relevant contraindications and precautions. Through such a knowledge graph, it can provide strong support for the research, teaching and clinical application of traditional Chinese medicine, greatly improving the accessibility and application efficiency of traditional Chinese medicine knowledge.

[0060] A pre-differentiation module uses expert experience knowledge graph hybrid retrieval to make preliminary disease differentiation and syndrome type judgment according to consultation needs.

[0061] The design of this module aims to assist doctors in quickly using existing expert experience and knowledge base to make preliminary analysis and evaluation of the patient's condition when facing complex cases, thereby providing a scientific basis for subsequent treatment plans.

[0062] The pre-differentiation module uses a hybrid retrieval strategy to retrieve relevant syndrome differentiation experience from the expert experience knowledge graph according to the actual disease, as the context of the large language model. The large language model is optimized for the specific needs of the TCM field, so that its output strictly follows the provisions of the "TCM Disease Syndrome Diagnosis and Treatment Criteria", ensuring that the output syndrome type expression is both accurate and in line with TCM professional terminology and expression habits.

[0063] The pre-differentiation module uses a hybrid retrieval strategy that combines multiple retrieval techniques to ensure that it can efficiently retrieve relevant syndrome differentiation experience from a TCM knowledge graph containing rich expert experience, such as Figure 4 The retrieved syndrome differentiation experience is then used as context information for the large language model. This large language model has been optimized specifically for the TCM field, so that its generated output strictly follows the provisions of the "TCM Disease Syndrome Diagnosis and Treatment Criteria", ensuring that the output syndrome type expression is both accurate and in line with TCM professional terminology and expression habits.

[0064] The expert experience knowledge graph is a system that integrates the disease differentiation and syndrome differentiation experience of various medical schools. It can achieve a series of complex functions. First, by using advanced large language model technology, the system can extract the diagnosis methods and comments of experts from tens of thousands of real medical records. Second, combined with the Graph RAG framework, the system can effectively extract key entities and their relationships in medical records, thereby constructing a structured knowledge network. Finally, the system integrates this information and generates representative community summaries, which are then written into the knowledge graph, providing strong data support for medical research and clinical practice.

[0065] The multi-agent diagnosis module is configured with multiple agents that collaborate in consultation based on the preliminary disease differentiation and syndrome type judgment. The multiple agents include:

[0066] The supervisor agent is responsible for coordinating and assigning consultation tasks to multiple expert agents, summarizing the diagnosis plans of multiple expert agents, and conducting individualized diagnosis based on the assigned consultation tasks and weighted retrieval of experience medical record data. The review agent performs risk verification of the TCM knowledge graph and optimization of the diagnosis plan to improve the diagnosis plan.

[0067] Each agent plays an indispensable role in the system, collectively forming this efficient collaborative consultation mechanism. The supervisory agent, as the command center of the system, ensures the efficient flow of consultation tasks through its precise coordination and dispatch capabilities. It can not only allocate tasks to various expert agents based on consultation needs and preliminary disease identification and syndrome type judgment results, but also effectively summarize the diagnosis plans of various experts after the consultation, providing a solid foundation for subsequent diagnosis optimization.

[0068] Expert agents each have their own knowledge base, which contains the valuable experience and wisdom of experts in various fields of traditional Chinese medicine. After receiving the tasks allocated by the supervisory agent, expert agents can quickly call on resources in their knowledge base, perform weighted retrieval of experience medical record data, and conduct personalized diagnosis. This process not only embodies the personalized characteristics of traditional Chinese medicine diagnosis and treatment, but also ensures the diversity and comprehensiveness of diagnosis plans.

[0069] The review agent plays the role of the last line of defense for diagnosis plans. Based on the traditional Chinese medicine knowledge graph, it performs risk verification and prescription optimization on the diagnosis plans submitted by expert agents. By constructing a traditional Chinese medicine contraindication graph, the review agent can accurately identify potential risks in diagnosis plans and replace contraindicated drugs, ensuring that the final diagnosis plan is both safe and effective.

[0070] The design of the multi-agent collaborative consultation system not only breaks through the limitations of traditional Chinese medicine diagnosis and treatment models, but also achieves the diversification and personalization of treatment methods. By introducing a multi-agent collaborative decision-making mechanism, it successfully simulates the multi-specialist collaboration process in real consultations, opening up new avenues for the intelligent development of traditional Chinese medicine.

[0071] As shown in Figure 2 Based on the above-mentioned multi-agent collaborative consultation system for traditional Chinese medicine famous prescriptions and techniques, the present application also discloses a multi-agent collaborative consultation method for traditional Chinese medicine famous prescriptions and techniques, comprising the following steps:

[0072] S1, constructing structured experience medical record data, expert experience knowledge graph, and traditional Chinese medicine knowledge graph knowledge base.

[0073] Structured experience medical record data is used to construct and enrich the knowledge base of the multi-agent collaborative consultation system for traditional Chinese medicine famous prescriptions and techniques. These data come from the accumulation of clinical practice, including the diagnosis process, treatment methods, used prescriptions, and efficacy feedback of various diseases. By structuring these data, key information such as symptom characteristics, prescription composition, drug dosage, and treatment period can be extracted, providing data support for subsequent agent decision-making. At the same time, structured experience medical record data also provides a basis for the continuous learning and optimization of the system, enabling the system to continuously absorb new clinical experience and knowledge, improving its diagnosis ability and treatment effect.

[0074] The expert experience knowledge graph knowledge base is used to store and manage expert experience knowledge, which comes from experienced experts in the field of traditional Chinese medicine, covering rich disease diagnosis experience, treatment strategies and prescription compatibility principles. Through the form of graph, the implicit knowledge of experts is made explicit, which facilitates the agent to quickly retrieve and apply in the consultation process. At the same time, this knowledge base also has the ability of dynamic updating, which can continuously absorb new expert experience and research results, ensuring that the knowledge base of the consultation system always maintains the forefront and accuracy.

[0075] The traditional Chinese medicine knowledge graph knowledge base is used to store and manage professional knowledge in the field of traditional Chinese medicine, which covers the nature, taste, meridian, efficacy, indication, usage, dosage and interaction relationship between drugs of Chinese herbal medicines. Through the form of graph, the complex knowledge system of traditional Chinese medicine is structured, so that the agent can quickly understand and apply these knowledge. The construction of traditional Chinese medicine knowledge graph knowledge base not only improves the professionalism of the consultation system, but also provides more accurate and comprehensive traditional Chinese medicine knowledge support for the agent in the consultation process, further enhancing the diagnostic ability of the system and the scientific nature of the treatment plan.

[0076] S2, according to the consultation demand, using expert experience knowledge graph mixed retrieval knowledge base, preliminary disease and syndrome type judgment pre-diagnosis.

[0077] In the pre-diagnosis stage, the present application adopts a comprehensive retrieval strategy. First, based on the specific symptoms of the patient, relevant syndrome experience is retrieved from a graph containing rich expert experience knowledge. This knowledge graph is carefully constructed and can provide diagnosis and treatment experience related to multiple symptoms. The retrieved syndrome experience serves as input information for the subsequent large language model, providing necessary context information. Subsequently, a large language model optimized for traditional Chinese medicine field knowledge is used to process this information. The model is trained and adjusted for traditional Chinese medicine field knowledge to ensure that its output results accurately conform to the syndrome type expressions specified in the "Standardization of Diagnosis and Treatment of Traditional Chinese Medicine Syndromes". Through this method, an accurate and traditional Chinese medicine diagnosis standard-compliant syndrome type description can be obtained, providing a scientific basis for subsequent treatment.

[0078] As Figure 4As shown, the construction of the expert experience knowledge graph focuses on integrating the disease and syndrome differentiation experience of multiple medical schools. The implementation process of this graph involves multiple steps. First, advanced large language model technology is used to analyze and learn tens of thousands of real medical records in depth, thereby summarizing the key experience and wisdom of experts in the diagnosis and review process. Next, the Graph RAG framework is used, which can effectively extract entity relationships in the text to help identify and understand various medical concepts involved in the medical records and their relationships. Finally, by generating community summaries, the extracted key information and entity relationships are written into the knowledge graph, thereby constructing a rich and accurate expert experience knowledge graph.

[0079] S3, according to the preliminary disease differentiation and syndrome type judgment of consultation, call multiple multi-agent for collaborative consultation, obtain risk verification, prescription optimization expert diagnosis scheme.

[0080] Among the multiple intelligent agents, the supervision intelligent agent is used to coordinate and assign diagnosis tasks to multiple expert intelligent agents; the expert intelligent agent is used to weighted retrieval of experience medical records as the main context of the model for further individualized diagnosis; the review intelligent agent is used for risk verification and prescription optimization based on the traditional Chinese medicine knowledge graph, which specifically includes the following steps:

[0081] S31, call the supervision intelligent agent to assign the consultation task to multiple expert intelligent agents;

[0082] S32, each expert intelligent agent performs individualized diagnosis and is summarized to the supervision intelligent agent to complete a round of diagnosis;

[0083] S33, when the predetermined diagnosis round is reached or the expert intelligent agent has no more questions, the supervision intelligent agent summarizes the expert diagnosis scheme and delivers it to the review intelligent agent;

[0084] S34, the review intelligent agent performs risk verification on the diagnosis scheme based on the traditional Chinese medicine knowledge graph and optimizes the diagnosis scheme.

[0085] The operation of the supervision intelligent agent includes: according to the consultation demand, selecting appropriate experts from the expert library to form an expert intelligent agent group, and sending independent diagnosis tasks to the members of the expert intelligent agent group. After completing a round of diagnosis, the supervision intelligent agent summarizes the diagnosis results of each expert intelligent agent and selects the subsequent processing flow according to the model context conversation.

[0086] Further, the supervision intelligent agent can ask the user to further supplement information in the form of questions. When the predetermined diagnosis round is reached or the expert sub-intelligent agent has no more questions, the supervision intelligent agent summarizes all expert diagnosis schemes and delivers them to the subsequent review intelligent agent.

[0087] The operation of the expert agent includes: accepting the diagnosis task assigned by the supervised agent, relying on the special knowledge set mounted by different expert agents, seeking differentiated and diversified diagnosis. Representative similar medical records are extracted from the experience medical record data knowledge base, and the experience medical record data is weighted searched as the context of the large language model RAG to encourage the model to refer to past successful cases when diagnosing.

[0088] When weighting searching experience medical record data, a similarity weighting summation algorithm is used to calculate the similarity of disease attribute embedding vectors, and a weight distribution graph is developed according to the difference between diseases. Key factors are considered when filtering medical records, such as: assigning a weight of 0.8-1.0 to core syndrome elements (such as looking at the spirit of stroke); Assign a weight of 0.5-0.7 to secondary elements, etc.

[0089] The operation of the review agent includes:

[0090] First, construct a contraindication relationship graph, extract medicinal material nodes, contraindication rule nodes and their associated edges from the traditional Chinese medicine knowledge graph, and construct a dynamically updated bipartite graph network. The edge weight between the medicinal material node and the contraindication node represents the contraindication strength.

[0091] Then, perform multi-dimensional conflict detection on the medicinal material set in the treatment plan to perform bidirectional breadth-first search.

[0092] Bidirectional breadth-first search includes: first direction: along the medicinal material to contraindication edge to detect the contraindication triggered by the patient's current state (such as pregnant women are prohibited from using safflower); Second direction: along the medicinal material to compatibility to contraindication edge to detect combination type contraindication (such as licorice against kudou).

[0093] When the total weight of the contraindication edge exceeds the preset threshold (such as ≥0.8), a conflict alert is generated; then the intelligent prescription optimization.

[0094] Intelligent prescription optimization can generate a replacement candidate set for the conflict medicinal material based on the medicinal material replacement relationship subgraph, and preferentially select medicinal materials with the same efficacy and no contraindications (such as motherwort replacing safflower); Equivalent medicinal materials with low contraindication strength (such as dried ginger replacing aconite with a dose of 0.8); At the same time, combine the patient's constitution data (yang deficiency / yin deficiency) to dynamically adjust the weight of the replacement scheme.

[0095] When a high-risk unannounced contraindication situation is identified (for example, the prescription contains musk ingredients and the patient does not provide pregnancy status information), a structured inquiry template is constructed, which should describe the specific situation of the contraindication conflict in detail and provide a comparison diagram of the alternative scheme. Based on the patient's feedback, update the contraindication detection results and perform interactive verification.

[0096] Application example:

[0097] For example As shown, input the patient's complaint and key four diagnostic information. Obtain the syndrome differentiation knowledge from the knowledge base (such as Figure 4 As shown) by the pre-diagnosis module, call the triage large model, and obtain the patient's disease type and syndrome type through detailed thinking chain analysis. In this process, the user will be dynamically interacted with to perfect the medical record information until it can support syndrome differentiation and further analysis.

[0098] Through the obtained disease type and syndrome type information, the RAG technology is used to filter the most skilled expert agent that matches the patient according to the description of the skilled field in the expert agent, and a consultation expert agent group is constructed. The expert agent group generally consists of 3-5 expert agents.

[0099] The supervisory agent packages the patient information and sends it to each expert agent for diagnosis. Each expert agent performs independent diagnosis, first queries the most similar historical medical records from the respective medical record knowledge base as analysis reference materials. During the diagnosis process, the problem is actively designed and sent back to the supervisory agent; if there is no problem, the review agent is used to analyze the specific diagnosis scheme, such as Chinese medicine prescription, acupuncture and massage, medical order, etc.

[0100] After completing a round of diagnosis, the supervisory agent collects possible expert questions (such as 5 as shown), summarizes the representative questions and interacts with the user, and then sends the user's answers to all expert agents. All expert agents share information and supplement, and perform the second round of diagnosis, and recursively until the diagnosis number limit is reached or the expert agent group has no more questions.

[0101] The supervisory agent comprehensively scores the final diagnosis of all expert agents, forms a summary report, and flows to the next risk review agent.

[0102] The review agent first filters the Chinese herbal medicine and prescription names used in the report, extracts the medicine node, taboo rule node and their associated edges from the traditional Chinese medicine knowledge graph, and constructs a bipartite graph network. Perform bidirectional breadth-first search, and screen for medicine-symptom conflict and medicine-compatibility conflict. Form a taboo risk suggestion report, or optimize the intelligent prescription, and finally deliver the expert diagnosis list with scores and risk suggestion report to the user.

[0103] The application also provides an electronic device, Figure 6 The structure schematic diagram of the electronic device provided by the embodiment of the application is as shown in Figure 6As shown, the electronic device may include a processor, a communications interface, memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions from the memory, for example, to execute the following method:

[0104] S1. Construct a knowledge base of structured medical case data, expert experience knowledge graph, and traditional Chinese medicine knowledge graph;

[0105] S2. Based on the consultation needs, use the expert experience knowledge graph to search the knowledge base to conduct preliminary disease identification and syndrome differentiation.

[0106] S3. Based on the preliminary diagnosis and syndrome differentiation made by the consultation, multiple multi-agents are invoked for collaborative consultation to obtain expert diagnostic solutions for risk verification and prescription optimization.

[0107] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] This invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments, including, for example:

[0109] S1. Construct a knowledge base of structured medical case data, expert experience knowledge graph, and traditional Chinese medicine knowledge graph;

[0110] S2. Based on the consultation needs, use the expert experience knowledge graph to search the knowledge base to conduct preliminary disease identification and syndrome differentiation.

[0111] S3. Based on the preliminary diagnosis and syndrome differentiation made by the consultation, multiple multi-agents are invoked for collaborative consultation to obtain expert diagnostic solutions for risk verification and prescription optimization.

[0112] The system embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0113] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0114] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-intelligent agent collaborative consultation system for traditional Chinese medicine, characterized in that, Comprise: Medical record knowledge base module, for collecting and storing diagnosis and treatment medical record, structured experience medical record data knowledge base is built; Expert knowledge base module, for collecting and storing famous expert syndrome differentiation logic and treatment scheme, structured expert experience knowledge graph knowledge base is built; Medicine knowledge base module, for extracting and storing traditional Chinese medicine prescription key attributes, structured traditional Chinese medicine knowledge graph knowledge base is built; Pre-differentiation module, according to the consultation demand, the expert experience knowledge graph is used to carry out preliminary disease and syndrome type judgment by mixed retrieval; Multi-agent diagnosis module, configured with multiple agents, according to the preliminary disease and syndrome type judgment of consultation, collaborative consultation is carried out.

2. The system of claim 1, wherein, The multiple agents include: Supervision agent, for coordinating and assigning consultation tasks to multiple expert agents; Summarize the diagnosis scheme of multiple expert agents; Multiple expert agents, according to the assignment of consultation task, weighted retrieval of experience medical record data is carried out for individualized diagnosis; Review agent, for risk checking of traditional Chinese medicine knowledge graph, optimization of diagnosis scheme.

3. A Chinese medicine name formula name technique multi-agent collaborative consultation method applied to the system of claim 1 or 2, characterized in that, Including steps: S1, structured experience medical record data, expert experience knowledge graph and traditional Chinese medicine knowledge graph knowledge base are built; S2, according to the consultation demand, the expert experience knowledge graph is used to retrieve the knowledge base by mixed retrieval, and preliminary disease and syndrome type judgment is carried out; S3, according to the preliminary disease and syndrome type judgment of consultation, multiple multi-agents are called to carry out collaborative consultation, and the expert diagnosis scheme of risk checking and prescription optimization is obtained.

4. The method of claim 3, wherein, The pre-differentiation in S2 includes steps: S21, according to the consultation disease, the mixed retrieval strategy is used to retrieve the relevant syndrome differentiation experience from the expert experience knowledge graph as the large language model prompt text; S22, based on the optimized large language model, the syndrome type expression conforming to the "standard of diagnosis and efficacy of traditional Chinese medicine" is output.

5. The method of claim 4, wherein, The expert experience knowledge graph includes: Using large language model to summarize expert diagnosis and comments from tens of thousands of real medical records, combining Graph RAG framework to extract entity relationship, generating community abstract and writing into knowledge graph.

6. The multi-agent collaborative consultation method for a traditional Chinese medicine prescription or formula according to claim 3, characterized in that, In S3, multiple multi-agents are called to carry out collaborative consultation, including steps: S31, call the supervision agent, assign the consultation task to multiple expert agents; S32, each expert agent carries out individualized diagnosis, and is summarized to the supervision agent to complete a round of diagnosis; S33, when reaching the predetermined diagnosis round or the expert agent has no more questions, the expert diagnosis scheme is summarized by the supervision agent and delivered to the review agent; S34, the review agent carries out risk checking based on the traditional Chinese medicine knowledge graph, and optimizes the diagnosis scheme.

7. The multi-agent collaborative consultation method for traditional Chinese medicine prescriptions and formulas according to claim 6, characterized in that, In S32, when the expert agent accepts the diagnosis task assigned by the supervision agent: Depending on different expert agents mounted with special knowledge base, and weighted retrieval of experience medical record data is carried out as the RAG prompt text of large language model, which stimulates the large language model to refer to experience medical record data in diagnosis.

8. The multi-agent collaborative consultation method of traditional Chinese medicine prescriptions and formulas according to claim 6, characterized in that, In S34, the risk checking and prescription optimization of diagnosis scheme of review agent includes: Based on the traditional Chinese medicine knowledge graph, a traditional Chinese medicine incompatibility relationship graph is constructed, wherein the edge weight between incompatibility nodes represents the incompatibility strength of medicinal materials, a search is performed on the medicinal material set in the diagnosis scheme to obtain incompatibility drugs, and the incompatibility drugs are replaced to optimize the diagnosis scheme.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the traditional Chinese medicine famous prescription and famous technique multi-agent collaborative consultation method according to any one of claims 3-8 when executing the program.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the traditional Chinese medicine famous prescription and famous technique multi-agent collaborative consultation method according to any one of claims 3-8.

Images