Multi-agent cooperative medical diagnosis method and device based on knowledge graph

Abstract

The invention discloses a multi-agent cooperative medical diagnosis method and device based on a knowledge graph, a storage medium and electronic equipment, and belongs to the technical field of artificial intelligence and medical information processing. In order to solve the technical problem of how to improve the reasoning ability, interpretability and diagnosis accuracy of a medical artificial intelligence diagnosis system in processing complex clinical tasks so as to assist diagnosis of doctors, the technical scheme adopted by the invention is as follows: (1) a multi-agent cooperative medical diagnosis method based on a knowledge graph; the method comprises the following steps: S1, a medical knowledge graph module; s2, a multi-agent cooperative diagnosis system module; and S3, generating a diagnosis report module. And (2) a multi-agent cooperative medical diagnosis device based on the knowledge graph, wherein the device comprises a medical knowledge graph construction unit, a multi-agent cooperative diagnosis system construction unit and a diagnosis report generation construction unit.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and medical information processing, and particularly to a multi-agent collaborative medical diagnosis method and device based on a knowledge graph. Background Art

[0002] In recent years, the emergence of large language models has led to a new round of innovation and development in artificial intelligence. The combination of large language models with traditional methods in many fields has been proven to have excellent performance. The traditional medical industry has also begun to integrate with large language models and achieved a series of remarkable achievements. Especially in medical knowledge Q&A, large language models have shown more powerful capabilities. Since large language models are pre-trained on a vast amount of general and medical texts, they have the ability to cover various medical knowledge such as basic medicine, clinical guidelines, drug information, case descriptions, etc. At the same time, the advantages of large language models in natural language understanding and generation enable them to accurately understand the semantics of questions and give answers with clear structures and fluent expressions. However, with the increase in the complexity of medical problems, large language models lacking good reasoning abilities cannot directly provide reasonable solutions. At the same time, current medical large models mostly operate as monomers, lacking task decomposition and collaboration mechanisms, making it difficult to achieve modular modeling and step-by-step reasoning of complex diagnostic processes. Moreover, medical reasoning highly relies on professional knowledge, and existing large language models often rely on their pre-trained corpora and are difficult to flexibly call external structured knowledge resources such as clinical guidelines and disease knowledge graphs, resulting in an increased risk of hallucinations. In this context, there is an urgent need for a collaborative diagnostic method that integrates a knowledge graph and a multi-agent architecture.

[0003] In recent years, multi-agent systems have gradually demonstrated unique advantages in various fields of artificial intelligence due to their collaborative reasoning capabilities in complex tasks. The multi-agent collaborative medical diagnosis method and device based on a knowledge graph improve the system's reasoning ability, knowledge expression ability, and diagnostic interpretability by introducing a collaborative working mechanism with clearly defined roles for agents, combined with the structured medical background provided by the knowledge graph, so as to better support highly reliable and highly complex clinical decision-making tasks. The present invention is expected to play an important role in multiple fields such as intelligent healthcare, chronic disease management, and hierarchical diagnosis and treatment, providing key support for building a safe, intelligent, and trustworthy medical AI system in the future. Summary of the Invention

[0004] In view of the deficiencies in the existing medical intelligent diagnosis methods, such as the uncontrollable reasoning process, insufficient knowledge invocation, and lack of interpretability of diagnosis results, the present invention proposes a multi-agent collaborative medical diagnosis method and device based on a knowledge graph. The method and device construct a diagnosis framework that integrates structured medical knowledge and agent division of labor and collaboration, which can simulate the clinical collaborative diagnosis process and jointly improve the accuracy and interpretability of diagnosis from the structured and semantic levels. The present invention first constructs a medical knowledge graph for diagnosis tasks by integrating a Chinese disease knowledge dataset; then introduces a multi-agent collaborative mechanism with role division, where different agents play different roles according to task requirements to complete subtasks such as triage, diagnosis, and knowledge graph retrieval. In each subtask, multiple agent proxies dynamically interact to obtain information through modular behavioral requirements; finally, the agent generates a multi-dimensional diagnosis report to achieve a complete closed-loop process from multi-source information understanding to structured output. Its core idea is to use the additional knowledge base of the knowledge graph to expand the knowledge reserve of the large language model, and improve the reasoning ability of the large language model through the communication of agents playing different roles, thereby improving the overall performance of the large language model in clinical diagnosis.

[0005] The technical task of the present invention is realized in the following way. A multi-agent collaborative medical diagnosis method based on a knowledge graph, which is composed of a medical knowledge graph module, a multi-agent collaborative diagnosis system module, and a generated diagnosis report module. It constructs a medical knowledge graph representing the correspondence between symptoms and diseases, uses multiple agents to complete the clinical diagnosis process, obtains the dialogue information of the diagnosis process, and then extracts key information from the dialogue information to obtain a multi-dimensional diagnosis report for using artificial intelligence agents to assist clinical diagnosis; specifically as follows:

[0006] S1. The medical knowledge graph module performs entity extraction and triple construction operations on the input Chinese disease knowledge dataset to obtain a medical knowledge graph;

[0007] S2. The multi-agent collaborative diagnosis system module: consists of a patient agent, a triage doctor agent, a knowledge graph retriever agent, a main doctor agent, a deputy doctor agent, and a medical examination agent. Through the form of multi-agent interaction, it completes the entire clinical diagnosis process and obtains dialogue information;

[0008] S3. The generated diagnosis report module inputs the dialogue information into the main diagnosis report generation member agent and the deputy diagnosis report generation member agent, and through the interaction of the main diagnosis report generation member agent and the deputy diagnosis report generation member agent, obtains a multi-dimensional diagnosis report.

[0009] Preferably, the construction process of the medical knowledge graph module is specifically as follows:

[0010] First, use the Chinese disease knowledge dataset, which contains various types of medical knowledge such as symptoms, diseases, and drugs. Through entity extraction operations, symptoms and diseases are extracted to form original symptom and disease entities. Subsequently, through the operation of constructing triples, disease-relation-symptom triples are formed. Select all disease-relation-symptom triples and remove duplicate entities to obtain a medical knowledge graph;

[0011] Preferably, the diagnostic process of the multi-agent collaborative diagnosis system module is specifically as follows:

[0012] The patient agent obtains the patient's personal information, processes the patient's personal information to generate the chief complaint and patient information, sends the patient information to the triage doctor agent to obtain the department information, and sends the chief complaint to the attending doctor agent and the knowledge graph retriever agent respectively. Obtain the diagnostic result from the attending doctor agent, which includes medical examination items. The patient agent sends the medical examination items to the medical examination agent to obtain the medical examination result; The knowledge graph retriever agent inputs the chief complaint of the patient agent, then extracts symptom entities from the chief complaint, sends the symptom entities to the medical knowledge graph for retrieving disease entities, obtains all retrieved disease entities, and then the knowledge graph retriever agent sorts the disease entities by frequency to obtain the knowledge graph feedback information, and sends the knowledge graph feedback information to the attending doctor agent; The attending doctor agent generates questions to interact with the deputy doctor, then generates the diagnostic result, and sends the diagnostic result to the patient agent to complete the entire diagnostic process and obtain the dialogue information;

[0013] Preferably, the construction process of the diagnostic report generation module is specifically as follows:

[0014] Input the dialogue information into the main diagnostic report generation member agent and the deputy diagnostic report generation member agent. Subsequently, the main diagnostic report generation member agent analyzes the dialogue information and extracts key information from it. If there are doubts about the extracted key information, questions are generated and sent to the deputy diagnostic report generation member agent. Subsequently, the deputy diagnostic report generation member agent gives suggestions and sends them to the main diagnostic report generation member agent; Finally, after the main diagnostic report generation member agent has no doubts, a multi-dimensional diagnostic report is obtained; The content of the multi-dimensional diagnostic report includes four parts: symptoms, relevant examination results, diagnostic results, and treatment suggestions.

[0015] The multi-agent collaborative medical diagnosis device based on the knowledge graph, as shown in the appendix Figure 2 shown, the device includes: a medical knowledge graph construction unit, a multi-agent collaborative diagnosis system construction unit, and a diagnostic report generation construction unit; respectively implement the functions of steps S1, S2, and S3 in the multi-agent collaborative medical diagnosis method based on the knowledge graph. The specific functions of each unit are as described below:

[0016] A medical knowledge graph construction unit is used to query disease entities in subsequent clinical diagnoses, extract symptoms, disease entities and their relationships from a Chinese disease knowledge dataset, form "symptom-relationship-disease" triples, and thus construct a medical knowledge graph;

[0017] A multi-agent collaborative diagnosis system construction unit is used to build a multi-agent collaborative working system that simulates the clinical diagnosis and treatment process; this unit consists of a patient agent, a triage doctor agent, a knowledge graph retrieval agent, a primary doctor agent, a secondary doctor agent, and a medical examination agent. Multiple agents cooperate with each other to complete the entire diagnosis process and obtain dialogue information;

[0018] A diagnostic report generation construction unit is used to integrate the key information in the dialogue information generated by the multi-agent collaborative diagnosis system construction unit. The primary diagnostic report generation agent and the secondary diagnostic report generation agent jointly analyze the dialogue information, and after reaching an agreement, output a multi-dimensional diagnostic report covering four parts: symptoms, relevant examination results, diagnosis results, and treatment suggestions.

[0019] A storage medium stores multiple instructions, and the instructions are loaded by a processor to execute the steps of the above-mentioned multi-agent collaborative medical diagnosis method based on a knowledge graph.

[0020] An electronic device includes: the above-mentioned storage medium; and a processor for executing the instructions in the storage medium.

[0021] The multi-agent collaborative medical diagnosis method and device based on a knowledge graph of the present invention have the following advantages:

[0022] (1) The present invention introduces a multi-agent collaborative mechanism driven by a large language model into the clinical auxiliary diagnosis task, constructs a structured intelligent diagnosis system, and significantly improves the reliability and scalability of complex medical task processing;

[0023] (2) The present invention introduces a knowledge graph as an external knowledge enhancement mechanism, which can effectively make up for the knowledge blind spots of the large language model and significantly reduce the risk of misjudgment caused by hallucinations during the diagnosis process;

[0024] (3) The present invention realizes task decomposition and role division by designing multi-role intelligent agents, can simulate the collaborative consultation process in real clinical practice, and is applicable to complex disease analysis and decision support;

[0025] (4) The diagnosis process in the present invention has traceability, and the behavior of each intelligent agent has context records and decision-making bases, which enhances the transparency and interpretability of the diagnosis suggestions and is convenient for doctors to review and adopt;

[0026] (5) The present invention generates multi-dimensional diagnostic reports, which can comprehensively summarize the entire diagnostic process, and the output content includes symptoms, relevant examination results, diagnostic results, and treatment suggestions, meeting the actual clinical use requirements;

[0027] (6) The knowledge graph constructed by the present invention has strong scalability, supports docking with various medical resources (guidelines, literature, electronic medical records), and is convenient for continuous updating and cross-institutional deployment;

[0028] (7) The diagnostic process of the present invention is highly consistent with the actual medical consultation process, has good practicability, and is applicable to various scenarios such as hospital auxiliary decision-making systems, intelligent consultation terminals, and telemedicine platforms;

[0029] (8) The intelligent agent role mechanism proposed by the present invention has high flexibility, can expand the agent types as needed (such as psychological assessment agents, insurance decision-making agents, etc.), and provides a basic framework for multi-field integration;

[0030] (9) The performance of the present invention has been significantly improved compared with previous methods, indicating that the present invention has strong practicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below with reference to the drawings.

[0032] Figure 1 It is a schematic structural diagram of a multi-agent collaborative medical diagnosis method based on a knowledge graph

[0033] Figure 2 It is a schematic diagram of the components of a multi-dimensional diagnostic report

[0034] Figure 3 It is a schematic diagram of a multi-agent collaborative medical diagnosis device based on a knowledge graph DETAILED DESCRIPTION OF THE INVENTION

[0035] The multi-agent collaborative medical diagnosis method and device based on a knowledge graph of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] Example 1:

[0037] The overall model framework structure of the present invention is as shown in the appendix Figure 1 shown. As shown in the appendix Figure 1It can be seen that the main technical framework of the present invention includes the following three modules, namely, the medical knowledge graph module, the multi-agent collaborative diagnosis system module, and the generated diagnosis report module. Among them, the disease and symptom information constitutes the Chinese disease knowledge data set and is sent to the medical knowledge graph module. The medical knowledge graph module takes the Chinese disease knowledge data set and the symptom entities obtained from the multi-agent collaborative diagnosis system module as inputs, and through entity extraction operations, obtains the original symptoms and disease entities, constructs triples, and obtains the disease-relationship-symptom triples. Subsequently, according to the entities and relationships in the disease-relationship-symptom triples, a medical knowledge graph is obtained and sent to the multi-agent writing diagnosis system module. The multi-agent collaborative diagnosis system module includes a patient agent, a triage doctor agent, a knowledge graph retriever agent, a medical examination agent, a primary doctor agent, and a secondary doctor agent; First, the patient agent takes the patient's personal information, the department information output by the triage doctor agent, the diagnosis result output by the primary doctor agent, and the medical examination result output by the medical examination agent as inputs; First, the patient agent sorts out the patient's personal information to obtain the patient information and sends it to the triage doctor agent; Subsequently, the patient agent summarizes the patient's personal information to obtain the chief complaint and sends it to the knowledge graph retriever agent and the primary doctor agent; Subsequently, the patient agent interacts with the primary doctor agent to obtain the medical examination items and sends them to the medical examination agent; Subsequently, the triage doctor agent takes the patient information output by the patient agent as input, analyzes the patient information, obtains the department information, and sends it to the patient agent; Subsequently, the knowledge graph retriever agent takes the chief complaint output by the patient agent and the disease entities output by the medical knowledge graph module as inputs, extracts keywords from the chief complaint to obtain symptom entities, and sends them to the medical knowledge graph module. At the same time, the disease entities are sorted according to the retrieval frequency to obtain the knowledge graph feedback information and send it to the primary doctor agent; Subsequently, the medical examination agent takes the medical examination information and the medical examination items output by the patient agent as inputs, matches them in the medical examination information, obtains the medical examination result and sends it to the patient agent; Subsequently, the primary doctor agent takes the chief complaint output by the patient agent, the knowledge graph feedback information output by the knowledge graph retriever agent, and the reply information output by the secondary doctor agent as inputs, diagnoses the chief complaint described by the patient agent, obtains the diagnosis result, and sends it to the patient agent; Subsequently, it asks the secondary doctor agent a question, obtains the question, and sends it to the secondary doctor agent; Through interaction with the patient agent, dialogue information is obtained and sent to the generated diagnosis report module; Finally, the secondary doctor agent takes the question output by the primary doctor agent as input, obtains the reply information through its own medical knowledge, and sends the reply information to the primary doctor agent.The diagnostic report generation module includes a secondary diagnostic report generation member agent and a primary diagnostic report generation member agent; the secondary diagnostic report generation member receives the conversation information obtained from the multi-agent collaborative diagnosis system module and the inquiries obtained from the primary diagnostic report generation member agent. For the inquiries, by analyzing the conversation information, it finds the key points missing in the inquiries from the conversation information, obtains suggestions, and sends the suggestions to the primary diagnostic report generation member agent; the primary diagnostic report generation member takes the conversation information obtained from the multi-agent collaborative diagnosis system module and the suggestions generated by the secondary diagnostic report generation member agent as inputs, extracts the key information in the conversation information, improves the extraction of the key information according to the suggestions generated by the secondary diagnostic report generation member agent. If there are still doubts, it generates inquiries and sends the inquiries to the secondary diagnostic report generation member agent. After extracting all the key information in the conversation information, a multi-dimensional diagnostic report is obtained as the output of the diagnostic report generation module.

[0038] Embodiment 2:

[0039] As shown in the appendix Figure 1 The multi-agent collaborative medical diagnosis method based on a knowledge graph of the present invention includes the following steps:

[0040] S1. Medical knowledge graph module: Extract the symptoms, disease entities and their relationships from the Chinese disease knowledge dataset to form "symptom-relationship-disease" triples, thereby constructing a medical knowledge graph. The specific steps are as follows:

[0041] S101. Extract entities: There are various types of knowledge such as symptoms, diseases, and drugs in the Chinese disease knowledge dataset. Retain the information in the symptom column and the disease column, and retain the relationship between disease and symptom. Take symptoms and diseases as two types of entities;

[0042] Example: Extract entities from the Chinese disease knowledge dataset. The example is as follows:

[0043] In the Chinese disease knowledge dataset, a disease has multiple aspects of description. Retain the disease column, that is, the "name" column, and at the same time retain the symptom column, that is, the "symptom" column. Extract the information in the keywords as entities, and let the relationship between disease and symptom be "has symptom", and the relationship between symptom and disease be "has disease".

[0044]

[0045]

[0046] S102. Construct triples: Input the symptom entity and the disease entity, and connect them through relationships to form a disease-relationship-symptom triple;

[0047] Example: Continuing to perform the operation of constructing triples on the data processed by S101 in the Chinese disease knowledge dataset, the example is as follows:

[0048] Correspond diseases with symptoms one by one to form multiple triples.

[0049]

[0050] S103. Construct a medical knowledge graph: Select all triples, merge all duplicate entities among them to form a medical knowledge graph;

[0051] Example: Continuing to perform the operation of constructing a medical knowledge graph on the triples constructed in S102 in the Chinese disease knowledge dataset, the example is as follows:

[0052] In the above triples, the disease entity "proximal fibular fracture" appears multiple times. After removing duplicates, only one disease entity is retained, and all associated symptom entities are connected to this disease entity, forming a connection between one disease entity and multiple symptom entities, thereby obtaining a medical knowledge graph.

[0053] S2. Multi-agent collaborative diagnosis system module: The multi-agent collaborative diagnosis system module is jointly composed of a patient agent, a triage doctor agent, a knowledge graph retriever agent, a primary doctor agent, a secondary doctor agent, and a medical examination agent;

[0054] S201. Patient agent: Obtain the patient's personal information, process the patient's personal information to generate the chief complaint and patient information, send the patient information to the triage doctor agent to obtain the department information; send the chief complaint to the primary doctor agent and the knowledge graph retriever agent respectively, obtain the diagnosis result from the primary doctor agent, which includes medical examination items, and send the medical examination items to the medical examination agent to obtain the medical examination result;

[0055] Example: The example of the patient's personal information is as follows:

[0056]

[0057] S202. Triage doctor agent: First, input the patient information output by the patient agent, then analyze the patient information to generate suitable department information, and thus send the department information to the patient agent for the patient agent to go to the corresponding department for treatment;

[0058] Example: Using the patient's personal information in S201, the patient agent interacts with the triage doctor agent, and the example is as follows:

[0059] The triage doctor agent recommends the corresponding department according to the patient's symptom description.

[0060] Patient Information Fever, chest tightness, cough, and difficulty breathing. Department Information I recommend you go to the internal medicine department for consultation.

[0061] S203. Knowledge Graph Retrieval Agent: First, input the chief complaint of the patient agent. Subsequently, extract symptom entities from the chief complaint. Then, send the symptom entities into the medical knowledge graph to retrieve disease entities, and obtain all the retrieved disease entities. Subsequently, the Knowledge Graph Retrieval Agent sorts the disease entities by frequency to obtain the knowledge graph feedback information. Finally, send the knowledge graph feedback information into the attending doctor agent;

[0062] Example: The Knowledge Graph Retrieval Agent extracts symptom entities such as "fever, muscle soreness" according to the chief complaint of the patient agent. Subsequently, use the symptom entities to retrieve in the medical knowledge graph to obtain disease entities such as "pneumonia". Subsequently, the Knowledge Graph Retrieval Agent selects the five disease entities with the highest frequency according to the retrieval frequency to obtain the knowledge graph feedback information "pneumonia, H7N9 avian influenza, pulmonary bullae, avian influenza."

[0063]

[0064] S204. Attending Doctor Agent: First, input the chief complaint generated by the patient agent and the knowledge graph feedback information. Combine the chief complaint and the knowledge graph feedback information to generate questions. Then, input the questions into the deputy doctor agent to obtain the reply information. Subsequently, generate a diagnosis result according to the diagnosis of the attending doctor agent, and send the diagnosis result into the patient agent;

[0065] Example: Using the patient's personal information in S201, the attending doctor agent interacts with the patient agent. The example is as follows:

[0066] First, the patient agent describes the symptoms through the chief complaint. The attending doctor agent analyzes the chief complaint to obtain the diagnosis result. At the same time, if medical examinations are required, the attending doctor will generate questions, input the questions into the deputy doctor agent to obtain the reply information, and then give the diagnosis result;

[0067]

[0068]

[0069] S205. Deputy Doctor Agent: Input the questions generated by the attending doctor agent. The deputy doctor agent analyzes the questions to obtain the reply information, and send the reply information into the attending doctor agent;

[0070] Example: In the example in S204, the part of the interaction between the deputy doctor agent and the attending doctor agent is described in detail. The example is as follows:

[0071]

[0072]

[0073] S206. Medical Examination Agent: Input medical examination items and medical examination information, match the medical examination items from the medical examination information to obtain the medical examination results, and send the medical examination results to the patient agent;

[0074] Example: In the example in S204, the part where the medical examination agent interacts with the patient agent is described in detail as follows:

[0075] The medical examination agent matches according to the medical examination items in the medical examination information and outputs the results in the medical examination information.

[0076]

[0077]

[0078] S3. Generate Diagnostic Report Template: First, input the conversation information obtained in S205 into the main diagnostic report generation member agent and the deputy diagnostic report generation member agent; then, the main diagnostic report generation member agent analyzes the conversation information and extracts key information from it. If there are doubts about the extracted key information, an inquiry is generated and sent to the deputy diagnostic report generation member agent. Subsequently, the deputy diagnostic report generation member agent gives suggestions and sends them to the main diagnostic report generation member agent; finally, after the main diagnostic report generation member agent has no doubts, a multi-dimensional diagnostic report is obtained. The specific steps are as follows:

[0079] S301: Load Conversation Information: Input the conversation information provided by the main doctor agent into the main diagnostic report generation member agent and the deputy diagnostic report generation member agent;

[0080] Example: A complete conversation information specifically includes the following:

[0081] Chief Complaint I am a 52-year-old female... Diagnosis Result Brucella Knowledge Graph Feedback Information Pneumonia, H7N9 avian influenza, pulmonary bulla, avian influenza. Reply Information …

[0082] S302: Interaction between the Main Diagnostic Report Generation Member Agent and the Deputy Diagnostic Report Generation Member Agent: The main diagnostic report generation member agent extracts key information from the conversation information. If there are doubts during the extraction process, an inquiry is generated and sent to the deputy diagnostic report generation member agent to obtain the suggestions summarized by the deputy diagnostic report generation member agent;

[0083] Example: Send the conversation information in S301 to the main diagnostic report generation member agent and the deputy diagnostic report generation member agent to enable the main diagnostic report generation member agent and the deputy diagnostic report generation member agent to interact. The example is as follows:

[0084] The main diagnostic report generation member agent and the deputy diagnostic report generation member agent interact on every detail of the conversation information, and the interaction ends after the opinions of the main diagnostic report generation member agent and the deputy diagnostic report generation member agent reach an agreement.

[0085]

[0086]

[0087] S303: Generate a multi - dimensional diagnostic report: If the main diagnostic report generation member agent has no doubts about extracting key information from the conversation information, the main diagnostic report generation member agent generates a multi - dimensional diagnostic report;

[0088] As shown in the Figure 2 appendix, the multi - dimensional diagnostic report includes four dimensions: symptoms, relevant examination results, diagnosis results, and treatment suggestions. Each dimension is further refined into multiple specific sub - dimensions. The specific explanations for each dimension are as follows:

[0089] (1). Symptoms: Include four parts: general information, chief complaint, current medical history, and past medical history:

[0090] General information: Includes basic information such as gender, age, occupation, etc., providing background support for individualized diagnosis;

[0091] Chief complaint: The most prominent discomfort actively reported by the patient, providing a preliminary clue for diagnosis;

[0092] Current medical history: Describes the onset time, development process, accompanying symptoms, etc. of the current disease, which is an important basis for understanding the evolution of the condition;

[0093] Past medical history: Records the patient's previous diseases, surgeries, medications, etc., helping to determine whether it is a recurrence of an old disease or a potential complication;

[0094] (2). Relevant examination results: Include physical examination and auxiliary examinations:

[0095] Physical examination: Includes physical sign information obtained by doctors through methods such as visual inspection, palpation, auscultation, etc., which is an important source of the initial clinical impression;

[0096] Auxiliary examinations: Include laboratory examinations, imaging examinations, etc., which are important means for disease confirmation;

[0097] (3). Diagnosis results: Include three parts: diagnosis, diagnostic basis, and differential diagnosis:

[0098] Diagnosis: The clear disease name, which is the core output of the entire report;

[0099] Diagnostic basis: Lists the evidence supporting the diagnosis, including symptoms, examination results, and knowledge graph reasoning paths, etc., to improve the interpretability of the report;

[0100] Differential diagnosis: Lists other possible diseases that are similar to the current disease but have been excluded, showing the doctor's comprehensive analysis process, which helps to reduce misdiagnosis and missed diagnosis;

[0101] (4) Treatment suggestions: Include part of the treatment plan:

[0102] Treatment plan: Covers drug treatment, surgical suggestions, lifestyle adjustments, etc.;

[0103] For example, a complete diagnostic report is as follows:

[0104]

[0105]

[0106]

[0107] The method proposed by the present invention can not only help doctors with auxiliary clinical diagnosis; but also enable medical students to interact with other intelligent agents in the method as doctors, creating a simulated clinical environment for medical students and helping to improve the clinical diagnosis ability of medical students; the model of the present invention has achieved advanced performance in the SDCM dataset, and the comparison of experimental results is shown in Table 1 specifically:

[0108] Table 1: Experimental results of the present method and other methods

[0109]

[0110] The present invention has been compared with existing methods and models, and the experimental results show that the method of the present invention has been greatly improved; among them, the first four rows are the experimental results of existing methods and models, and the last row is the experimental result of the model of the present invention.

[0111] Example 3:

[0112] As shown in the appendix Figure 3 A multi-agent collaborative medical diagnosis device based on a knowledge graph, the device includes: a medical knowledge graph construction unit, a multi-agent collaborative diagnosis system construction unit, and a generated diagnostic report construction unit; respectively, based on the functions of steps S1, S2, and S3 in the multi-agent collaborative medical diagnosis method based on a knowledge graph first, the specific functions of each unit are as follows:

[0113] The medical knowledge graph construction unit is used to query disease entities in subsequent clinical diagnoses, extract "symptom-disease" medical entities and their semantic relationships from the Chinese disease knowledge dataset, generate standardized triples, and store them in a graph database to form a queryable medical knowledge graph;

[0114] The multi-agent collaborative diagnosis system construction unit is used to build an intelligent agent collaborative working system that simulates the clinical diagnosis and treatment process; this unit consists of a patient agent, a triage doctor agent, a knowledge graph retrieval agent, a main doctor agent, an associate doctor agent, and a medical examination agent, and multiple agents cooperate with each other to complete the entire diagnosis process;

[0115] Generate a diagnostic report construction unit for integrating key information in the multi-agent collaborative diagnostic system construction unit. The main diagnostic report generation agent and the secondary diagnostic report generation agent jointly analyze the dialogue information. After reaching an agreement, a multi-dimensional diagnostic report covering four parts: symptoms, relevant examination results, diagnostic results, and treatment suggestions is output.

[0116] Example 4:

[0117] Based on the storage medium of Example 2, which stores multiple instructions. The instructions are loaded by a processor to execute the steps of the knowledge graph-based multi-agent collaborative medical diagnostic method of Example 2.

[0118] Example 5:

[0119] An electronic device based on Example 4, the electronic device includes: the storage medium of Example 4; and a processor for executing the instructions in the storage medium of Example 4.

[0120] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

[0121] In this case, the program code read from the storage medium itself can implement the functions of any one of the above embodiments. Therefore, the program code and the storage medium storing the program code constitute a part of the present invention.

[0122] Embodiments of the storage medium for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RYM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Optionally, the program code can be downloaded from a server computer via a communication network.

[0123] In addition, it should be clear that not only can the functions of any one of the above embodiments be realized by executing the program code read by the computer, but also by the operating system or the like operating on the computer based on the instructions of the program code to complete part or all of the actual operations.

[0124] In addition, it can be understood that the program code read out from the storage medium is written into the memory provided in the expansion board inserted into the computer or into the memory provided in the expansion unit connected to the computer, and then based on the instructions of the program code, the CPU or the like installed on the expansion board or the expansion unit is made to execute part or all of the actual operations, thereby realizing the functions of any one of the above embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-agent collaborative medical diagnosis method based on a knowledge graph, characterized in that This method is a multi-agent collaborative medical diagnosis model composed of a medical knowledge graph module, a multi-agent collaborative diagnosis system module, and a diagnosis report generation module. It simulates the real clinical diagnosis process and can simulate the reasoning and decision-making processes of complex medical tasks to achieve the goal of medical artificial intelligence assisting clinical diagnosis, as follows: S1. Medical knowledge graph module: First, construct a medical knowledge graph for retrieval. Perform entity extraction on the Chinese disease knowledge dataset, extract each disease and its corresponding various symptoms to form original symptom and disease entities. Subsequently, construct triples to form disease-relationship-symptom triples, and then construct a medical knowledge graph connecting diseases and symptoms; S2. Multi-agent collaborative diagnosis system module: Composed of a patient agent, a triage doctor agent, a knowledge graph retrieval agent, a primary doctor agent, a secondary doctor agent, and a medical examination agent. Among them, the patient agent obtains the patient's personal information, processes the patient's personal information to generate the chief complaint and patient information, sends the patient information to the triage doctor agent to obtain the department information, and sends the chief complaint to the primary doctor agent and the knowledge graph retrieval agent respectively. Obtain the diagnosis result from the primary doctor agent, which includes medical examination items, and send the medical examination items to the medical examination agent to obtain the medical examination result; The knowledge graph retrieval agent inputs the chief complaint of the patient agent, then extracts symptom entities from the chief complaint, and then sends the symptom entities to the medical knowledge graph for retrieving disease entities to obtain all retrieved disease entities. Subsequently, the knowledge graph retrieval agent sorts the disease entities by frequency to obtain knowledge graph feedback information, and then sends the knowledge graph feedback information to the primary doctor agent; The primary doctor agent generates questions to interact with the secondary doctor, then generates the diagnosis result, and sends the diagnosis result to the patient agent to complete the entire diagnosis process and obtain the dialogue information; S3. Generate diagnosis report module: Interact through the primary diagnosis report generation member agent and the secondary diagnosis report generation member agent to finally generate a multi-dimensional diagnosis report; The content of the multi-dimensional diagnosis report includes four parts: symptoms, relevant examination results, diagnosis results, and treatment suggestions.

2. The multi-agent collaborative medical diagnosis method based on a knowledge graph according to claim 1, wherein The medical knowledge graph module, the specific implementation process is as follows: S1. Medical knowledge graph module: Extract symptoms, disease entities and their relationships from the Chinese disease knowledge dataset to form "symptom-relationship-disease" triples, thereby constructing a medical knowledge graph. The specific steps are as follows: S101. Entity extraction: There are various types of knowledge such as symptoms, diseases, and drugs in the Chinese disease knowledge dataset. Retain the information in the symptom column and the disease column, and retain the relationship between diseases and symptoms. Regard symptoms and diseases as two types of entities; S102. Construct triples: Use the symptom and disease entities as the two entities in the triples, and their relationship as the edge in the triples to form a disease-relationship-symptom triple; S103. Construct a medical knowledge graph: Select all triples, merge all duplicate entities among them to form a medical knowledge graph.

3. The multi-agent collaborative medical diagnosis method based on a knowledge graph according to claim 1, characterized in that The multi-agent collaborative diagnosis system module, the specific implementation process is as follows: S2. Multi-Agent Collaborative Diagnosis System Module: The multi-agent collaborative diagnosis system module is jointly composed of a patient agent, a triage doctor agent, a knowledge graph retriever agent, a primary doctor agent, a secondary doctor agent, and a medical examination agent. Through the interaction of multiple agents, the clinical diagnosis of the patient agent is completed; S201. Patient Agent: Obtain the patient's personal information, process the patient's personal information to generate the chief complaint and patient information, send the patient information to the triage doctor agent to obtain the department information; send the chief complaint to the primary doctor agent and the knowledge graph retriever agent respectively, obtain the diagnosis result from the primary doctor agent, which includes medical examination items, and send the medical examination items to the medical examination agent to obtain the medical examination result; S202. Triage Doctor Agent: First input the patient information output by the patient agent, then analyze the patient information to generate the department information, and then send the department information to the patient agent for the patient agent to go to the corresponding department for treatment; S203. Knowledge Graph Retriever Agent: First input the chief complaint of the patient agent, then extract the symptom entities from the chief complaint, then send the symptom entities to the medical knowledge graph for retrieving disease entities, obtain all retrieved disease entities, then the knowledge graph retriever agent sorts the disease entities by frequency to obtain the knowledge graph feedback information, and finally send the knowledge graph feedback information to the primary doctor agent; S204. Primary Doctor Agent: First input the chief complaint generated by the patient agent and the knowledge graph feedback information, combine the chief complaint and the knowledge graph feedback information to generate questions, then send the questions to the secondary doctor agent to obtain the reply information, and then generate the diagnosis result according to the diagnosis of the primary doctor agent, and send the diagnosis result to the patient agent; S205. Secondary Doctor Agent: Input the questions generated by the primary doctor agent, the secondary doctor agent analyzes the questions to obtain the reply information, and send the reply information to the primary doctor agent; S206. Medical Examination Agent: Input the medical examination items and medical examination information, match the medical examination items from the medical examination information to obtain the medical examination result, and send the medical examination result to the patient agent.

4. The multi-agent collaborative medical diagnosis method based on a knowledge graph according to claim 1, wherein Diagnosis Report Generation Module, specifically as follows: S3. Generate Diagnosis Report Template: First input the dialogue information into the primary diagnosis report generation member agent and the secondary diagnosis report generation member agent. Through the interaction of the primary diagnosis report generation member agent and the secondary diagnosis report generation member agent, obtain a multi-dimensional diagnosis report; S301. Load Dialogue Information: Input the dialogue information provided by the primary doctor agent into the primary diagnosis report generation member agent and the secondary diagnosis report generation member agent; S302. Interaction between the Primary Diagnosis Report Generation Member Agent and the Secondary Diagnosis Report Generation Member Agent: The primary diagnosis report generation member agent extracts the key information in the dialogue information. If there are any doubts during the extraction process, generate inquiries and send the inquiries to the secondary diagnosis report generation member agent to obtain the suggestions summarized by the secondary diagnosis report generation member agent; S303. Generate a Multi-Dimensional Diagnosis Report: If the primary diagnosis report generation member agent has no doubts when extracting the key information from the dialogue information, the primary diagnosis report generation member agent generates a multi-dimensional diagnosis report.

5. A multi-agent collaborative medical diagnosis device based on a knowledge graph, characterized in that, The medical knowledge graph construction unit, the multi-agent collaborative diagnosis system construction unit, and the diagnostic report generation construction unit respectively implement the knowledge graph-based multi-agent collaborative medical diagnosis method described in claims 1-4, as follows: The medical knowledge graph construction unit is used to query disease entities in subsequent clinical diagnoses, extract entities and their semantic relationships from the Chinese disease knowledge dataset, generate disease-relationship-symptom triples, and combine all disease-relationship-symptom triples to form a medical knowledge graph; The multi-agent collaborative diagnosis system construction unit is used to build an intelligent agent collaborative working system that simulates the clinical diagnosis and treatment process; this unit consists of a patient agent, a triage doctor agent, a knowledge graph retriever agent, a primary doctor agent, a secondary doctor agent, and a medical examination agent, and multiple agents cooperate with each other to complete the entire diagnosis process; The diagnostic report generation construction unit is used to integrate the key information in the multi-agent collaborative diagnosis system construction unit. The primary diagnostic report generation agent and the secondary diagnostic report generation agent jointly analyze the dialogue information, and after reaching an agreement, output a multi-dimensional diagnostic report covering four parts: symptoms, relevant examination results, diagnosis results, and treatment suggestions.

6. An electronic device, characterized in that, Including: A memory and at least one processor; Wherein, a computer program is stored on the memory; The at least one processor executes the computer program stored in the memory, so that the at least one processor executes the knowledge graph-based multi-agent collaborative medical diagnosis method described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and the computer program can be executed by a processor to implement the knowledge graph-based multi-agent collaborative medical diagnosis method described in any one of claims 1 to 4.

Images