Urinary surgery pre-inquiry method and system based on multi-agent cooperation

Abstract

The invention discloses a urinary surgery pre-inquiry method and system based on multi-agent cooperation. The system is driven by a node-routing graph, a dialogue state object is adopted to penetrate through the whole interrogation process, and the system runs according to the workflow of dialogue information collection, case characteristic manuscript formation, integrity auditing, patient confirmation and arbitration and first disease course record generation. The dialogue collection agent guides to obtain information such as chief complaint and accompanying symptoms; the expert manuscript forming agent generates and iteratively perfects structured fields such as current medical history, past history, physique and auxiliary examination, and outputs a control instruction for minimizing a question; an integrity auditing agent performs structure and content verification according to the specialized key point list, and loop inquiry is performed when the content is insufficient; an arbitration agent identifies patient feedback and drives a revised closed loop. The system is internally provided with a hybrid retrieval and rearrangement module, BM25 and vector semantic matching are fused, Top-K evidence-based literatures are selected by combining cross encoder scoring and Z-score normalization, an expert diagnosis agent generates a diagnosis basis with reference clues, preliminary diagnosis, differential diagnosis and diagnosis and treatment plans according to the Top-K evidence-based literatures, and a normalized first disease course record is output. According to the scheme, structured collection, automatic manuscript formation and closed-loop auditing of the inquiry information are achieved, traceable evidence-based support and patient participation are considered, the medical record quality and the doctor working efficiency are remarkably improved, and good expandability and clinical application value are achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of artificial intelligence and medical informatization technology, and specifically relates to a urinary surgery pre-consultation method and system based on multi-agent cooperation. BACKGROUND

[0002] With the continuous development of artificial intelligence technology and large language models (LLM), intelligent pre-consultation, automated medical records, and clinical decision support in the medical field are becoming a hot research and application direction. By introducing intelligent question answering, generative models, and structured data processing technology, it is expected to significantly improve the efficiency of medical information collection and the quality of medical record generation.

[0003] However, existing online consultation systems and electronic medical record platforms still face many challenges in practical application. At present, most systems can achieve information collection through patient free-text input, but there are still deficiencies in structured processing and context understanding, making it difficult to directly map the obtained information to standardized medical record templates, resulting in doctors still needing to invest a lot of effort in sorting and supplementing. In addition, clinical diagnosis and treatment highly depend on evidence-based medical guidelines and literature support, but existing systems generally lack efficient and traceable evidence-based retrieval and recommendation capabilities, making it difficult to provide doctors with accurate literature references in a timely manner, thereby affecting the scientificity and consistency of diagnosis and decision-making. More importantly, some generative models still have problems of uncertainty and insufficient boundary control in outputting content, which may introduce factual errors and even bring risks of medical compliance and quality control.

[0004] Therefore, there is an urgent need for an intelligent consultation system with structured output control, closed-loop audit mechanism, and traceable evidence-based support capability. The system should achieve information collection, structured drafting, intelligent audit, patient confirmation, and evidence-based suggestion generation through natural language interaction in multiple stages of intelligent collaboration process, and finally output high-quality, compliant, and usable medical record texts and diagnosis and treatment suggestions, thereby significantly reducing the information collection burden of doctors and improving the quality and traceability of medical records. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application provides a urinary surgery pre-consultation method and system based on multi-agent cooperation.

[0006] The present application introduces a node-routing graph driven multi-agent collaboration mechanism, structured dialogue state object management, closed-loop audit and patient confirmation process, and traceable evidence-based recommendation strategy, thereby realizing information collection structuring, case drafting standardization, audit revision automation, and diagnosis and treatment suggestion personalization throughout the consultation process, solving the problems of insufficient information structuring, difficult audit revision, and lack of evidence-based support in existing online consultation systems.

[0007] The first aspect of the application relates to a urinary surgery pre-consultation method based on multi-agent cooperation, comprising the following steps: Dialogue information collection: the system collects natural language interaction between the agent and the patient through dialogue, automatically guides the collection of patient basic information, chief complaint, accompanying symptoms and other content, and structures and records them in the dialogue state object.

[0008] Case feature writing: the expert writing agent extracts and fills in the case feature fields including present illness history, past history, physical examination, etc. according to the dialogue information, and generates subsequent interrogation control instructions to drive the system to continue to collect supplementary information.

[0009] Integrity audit: the integrity audit agent checks the structure and content integrity of the case features according to the urinary surgery specialty points, and returns the minimum follow-up instructions when it is incomplete, forming a local closed-loop mechanism of collection-writing-audit.

[0010] Patient confirmation and arbitration: the system displays the preliminary writing content to the patient, and the arbitration agent identifies the patient's feedback type. If there are revision opinions, the writing and auditing process will be iterated again to ensure patient participation and accuracy of medical records.

[0011] First course record generation: the expert diagnosis agent calls the hybrid retrieval and rearrangement module to complete literature recall and precision sorting based on the case features, generates diagnosis basis with evidence-based references, preliminary diagnosis, differential diagnosis and treatment plan, and outputs a complete and structured first course record.

[0012] Further, the node-routing graph mechanism adopted by the system clearly defines the state transfer path between agents, making the collaboration between modules orderly and highly decoupled.

[0013] Further, the dialogue state object is used for context recording and updating throughout the entire consultation process, with advantages such as field uniformity, data traceability, and transparent interaction.

[0014] Further, the hybrid retrieval and rearrangement module combines BM25 retrieval and vector semantic matching, supplemented by cross-encoder and Z-score normalization method, to achieve high-quality and accurate evidence-based literature recommendation.

[0015] The second aspect of the application relates to a urinary surgery pre-consultation system based on multi-agent cooperation, which is driven by node-routing graph to cooperate with multiple agents. Each module acts as an independent agent node, and its state transfer is controlled by predefined routing conditions. The dialogue state object runs through the entire system process and is used for unified recording, transmission and updating of interactive information and control instructions. It comprises: The dialogue information collection module is used for collecting basic information, chief complaint, accompanying symptoms and the like of the patient through natural language interaction, and structuring and storing the information in a dialogue state object; The case feature writing module is used for extracting and filling in the present illness history, past history, physical examination and the like based on the collected information, and generating an inquiry control instruction; The integrity auditing module is used for automatically detecting the information integrity of the medical record structure according to the urology specialty points, and generating a minimum follow-up instruction to return to the collection process if the information is not complete; The patient confirmation and arbitration module is used for showing the preliminary writing to the patient and identifying the feedback type, and responding to confirmation, revision or questions and driving the revision process; The first course record generation module calls the mixed retrieval and rearrangement module based on the case features to retrieve and screen the literature, and generates a diagnosis basis containing evidence-based references, a preliminary diagnosis, a differential diagnosis and a diagnosis and treatment plan.

[0016] The node-routing graph defines the transfer path and trigger condition between the intelligent agents, realizes the closed-loop control and collaborative decoupling of the process.

[0017] The dialogue state object includes the following fields: message record, inquiry round, inquiry control instruction, case feature, diagnosis basis, preliminary diagnosis, differential diagnosis, diagnosis and treatment plan and specific revision points.

[0018] The mixed retrieval and rearrangement module combines the BM25 keyword retrieval and vector semantic matching mechanism, and is supplemented by cross-encoder scoring and Z-score normalization, to realize the precise recommendation of high-quality evidence-based literature.

[0019] The integrity auditing module uses a minimum follow-up method to guide the system to collect missing information according to the preset urology points list, and realizes a multi-round iterative completion mechanism.

[0020] The arbitration module uses a semantic recognition model to judge the patient feedback type, automatically extracts revision points and drives the backflow to the writing and auditing process.

[0021] The system adapts to the structured format of the electronic medical record, and generates content to support subsequent clinical decision-making, teaching review and medical knowledge base construction.

[0022] The intelligent agent modules are highly decoupled, support cross-department migration deployment, and can be expanded to other specialty fields by adjusting the auditing points and knowledge base.

[0023] The beneficial effects of the present application are as follows: 1. Realize the whole process of structured and closed-loop diagnosis. The invention realizes the closed-loop workflow from information collection, case drafting, integrity audit, patient confirmation to the first course record generation by combining node-routing graph and dialogue state object. Each link is completed by independent agent cooperation, and data is dynamically updated in a unified structure, ensuring the structured, integrity and consistency of the diagnosis content, and greatly reducing the doctor's later work load.

[0024] 2. Ensure the standardization and traceability of medical record drafting. The expert drafting agent generates and completes the content according to the specialty template and standardized field, so that the medical record text meets the electronic medical record standard in structure and language. All question and answer records, revision points and system instructions are stored in the dialogue state object, forming an auditable and traceable whole process data chain.

[0025] 3. Build a multi-agent collaboration mechanism to realize division of labor and decoupling. The system adopts a multi-agent architecture driven by node-routing graph, and each agent has clear division of labor and is independent in dialogue collection, drafting, auditing, arbitration and diagnosis generation, realizing modular extension and flexible combination, and having good maintainability and scalability.

[0026] 4. Introduce automatic audit and patient confirmation mechanism to improve data quality. The integrity audit agent automatically checks the case structure and content integrity according to the urology key points list, and triggers the minimum follow-up loop; the arbitration agent performs semantic recognition and revision driving on patient feedback, forming a "system-patient-auditor" three-party closed loop, ensuring the accuracy of medical records and patient participation.

[0027] 5. Integrate evidence-based recommendation module to enhance the scientific nature of diagnosis. The mixed retrieval and rearrangement module combines BM25 and vector semantic retrieval, and uses cross-encoder scoring and Z-score normalization strategy to realize high-quality evidence-based literature recall and precision sorting. The expert diagnosis agent generates diagnosis basis and treatment plan with reference clues based on selected literature, making the suggestions more scientific, transparent and traceable.

[0028] 6. Support clinical teaching and knowledge sedimentation. The system stores structured diagnosis data while sedimenting case characteristics and diagnosis reasoning process into knowledge units, providing support for clinical training, case review and medical knowledge base construction, and promoting the continuous optimization of intelligent medical system.

[0029] 7. Good generality and expansion value Although the urology department is used as an example, the system framework, node-routing mechanism and multi-agent collaboration logic can be migrated to other specialty fields, and can be quickly adapted to different disease scenarios by adjusting the knowledge base and audit points, having wide clinical promotion and industrialization prospects. BRIEF DESCRIPTION OF DRAWINGS

[0030] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0031] Figure 1 This is an overall flowchart of the urological pre-diagnosis system based on multi-agent collaboration of the present invention.

[0032] Figure 2 This is the node-routing graph described in this embodiment of the invention.

[0033] Figure 3 This is a flowchart of the hybrid retrieval and rearrangement process in an embodiment of the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0035] Example 1

[0036] This embodiment proposes a urological pre-diagnosis method based on multi-agent collaboration. For example... Figure 2 As shown, a multi-agent workflow is driven by a node-routing graph and dialogue state objects. The node-routing graph consists of nodes and routes, where nodes correspond to different agents, and routes define the state transition conditions between nodes. Solid lines in the figure represent unconditional edges, and dashed lines represent conditional edges; state transitions in the system are triggered by nodes based on routing rules.

[0037] The dialogue state object is a structured data unit used to record, transmit, and update context information during system operation. It contains a set of fixed fields for storing information such as message records, number of consultation rounds, consultation control instructions, case characteristics, diagnostic basis, preliminary diagnosis, differential diagnosis, treatment plan, and specific revision points.

[0038] This system achieves closed-loop management of pre-consultation through a complete process from dialogue information collection, case drafting, review and confirmation to the generation of evidence-based treatment recommendations, thereby significantly improving the structure, information completeness, and evidence-based traceability of the consultation process. This method includes the following steps: Step 1: Collecting Dialogue Information In the present application, the dialogue collection agent is a modular unit specially used for collecting natural language information of patients. The system first initiates the questioning actively through the agent, and guides the patient to describe the disease in natural language. During the diagnosis process, the system determines whether it is the first round of diagnosis through the diagnosis round field in the dialogue state object. If it is the first round, the system will ask questions according to the preset standardization first round of diagnosis template of urology surgery, and collect the basic information of the patient such as gender, age, main symptoms, disease duration and accompanying symptoms; if it is subsequent diagnosis round, the questioning is performed according to the diagnosis control instruction in the dialogue state object, which is maintained and updated by the expert drafting agent.

[0039] During the whole process, all the question and answer contents are recorded in the message record field of the dialogue state object, providing a unified and structured data input basis for the subsequent case drafting, integrity audit and patient confirmation link.

[0040] Step 2: Case feature drafting

[0041] After the dialogue collection agent completes the questioning, the patient answers according to the system prompt, and the answer content will be stored in the message record field of the dialogue state object. Subsequently, the expert drafting agent reads the patient's answer content, and corrects, completes or rewrites the case feature field in the dialogue state object. The case features include patient basic information, present illness history, past medical history, physical examination and auxiliary examination, which are used to fully reflect the characteristics of the patient's condition.

[0042] At the same time, the dialogue collection agent generates new diagnosis control instructions according to the message record in the dialogue state object and the updated case feature field, and writes them into the dialogue state object. The diagnosis control instruction is used to guide the subsequent diagnosis content and questioning path, so as to realize the dynamic and closed collection dialogue process.

[0043] Step 3: Integrity audit

[0044] After a round of case feature drafting is completed, the system calls the integrity audit agent under the driving of the node-routing graph to audit the current case feature text. The audit is based on the list of urology specialty points, and checks the structural integrity of the case content and the sufficiency of the key information item by item.

[0045] If the audit result shows that the information is missing or incomplete, the integrity audit agent will generate the minimum follow-up instruction, and update the instruction to the dialogue state object, and the system will return to step 1 under the driving of the node-routing graph to continue the targeted diagnosis supplement. If the audit is passed and marked as "COMPLETE", it means that the case structure is complete and the information is sufficient, and the system will automatically enter the next stage of patient confirmation process.

[0046] Through the above mechanism, the system forms a local closed loop between "step 1-2-3", supports multi-round and step-by-step completion type information collection process, and thus ensures the completeness and high-quality of the case record.

[0047] Step 4: Patient confirmation and arbitration

[0048] When the case feature text is confirmed to be complete, the system will display the text to the patient and prompt the patient to check and confirm. The natural language feedback of the patient is recorded in the dialogue state object and read and analyzed by the arbitration intelligent agent. The arbitration intelligent agent will automatically determine the type of the patient's reply as "confirmation", "revision" or "other questions".

[0049] If the patient confirms that there is no error, the system will enter step 5; if the patient proposes a modification, the arbitration intelligent agent will identify the specific revision points from the patient's reply and update them to the dialogue state object. At this time, under the driving of the node-routing graph, the expert writing intelligent agent will read the revision points and the original case features, and make targeted revisions to the case content according to the patient's feedback. After the revision is completed, the node-routing graph will drive the integrity audit intelligent agent to review the revised text again.

[0050] If the patient has other questions, the system will jump back to the dialogue collection intelligent agent under the control of the node-routing graph and continue the interactive question and answer.

[0051] Through this design, the system can fully respond to patient feedback and form a closed-loop confirmation and revision mechanism, thereby further improving user experience and the accuracy and reliability of case data.

[0052] Step 5: First medical record generation

[0053] After the patient confirms the case features, the system starts the first medical record generation process under the driving of the node-routing graph. The first medical record content includes case features, diagnosis basis, preliminary diagnosis, differential diagnosis and treatment plan, which is generated by the expert diagnosis intelligent agent. The agent takes the case features as the core input, automatically reasons and generates the corresponding diagnosis and treatment plan.

[0054] Embodiment 2

[0055] As Figure 1 , the present embodiment relates to a urology pre-diagnosis system based on multi-agent cooperation, which is driven by node-routing graph to cooperate with multiple agents, each module serving as an independent agent node, and its state transition is controlled by pre-defined routing conditions; The dialogue state object runs through the entire system process and is used to uniformly record, transfer and update interactive information and control instructions; including: The dialogue information collection module is configured to collect basic information, chief complaint, accompanying symptoms, and the like of the patient through natural language interaction, and to store the information in a structured manner in a dialogue state object; The case feature generation module is configured to extract and fill in the present illness history, past medical history, physical examination, and the like based on the collected information, and to generate an inquiry control instruction; The integrity auditing module is configured to automatically detect the information integrity of the medical record structure according to the urology specialty points, and to generate a minimum follow-up instruction to return to the collection process if the information is incomplete; The patient confirmation and arbitration module is configured to show the preliminary draft to the patient and identify the feedback type, and to respond to confirmation, revision, or questions and drive the revision process; The first course record generation module is configured to call the mixed retrieval and rearrangement module based on the case features to perform literature recall and screening, and to generate a diagnosis basis containing evidence-based references, a preliminary diagnosis, a differential diagnosis, and a diagnosis and treatment plan, as shown in Figure 3

[0056] The node-routing graph defines the transfer path and trigger condition between the intelligent agents, realizes closed-loop control and collaborative decoupling of the process, as shown in Figure 2

[0057] The dialogue state object includes the following fields: message record, inquiry round, inquiry control instruction, case feature, diagnosis basis, preliminary diagnosis, differential diagnosis, diagnosis and treatment plan, and specific revision points.

[0058] The mixed retrieval and rearrangement module combines BM25 keyword retrieval and vector semantic matching mechanism, supplemented by cross-encoder scoring and Z-score normalization, to realize precise recommendation of high-quality evidence-based literature.

[0059] The integrity auditing module uses a minimum follow-up method to guide the system to collect missing information according to the preset urology points list, and realizes a multi-round iterative completion mechanism.

[0060] The arbitration module uses a semantic recognition model to determine the patient feedback type, automatically extracts revision points, and drives the flow back to the draft and auditing process.

[0061] The system adapts to the structured format of the electronic medical record, and the generated content supports subsequent clinical decision-making, teaching review, and medical knowledge base construction.

[0062] The intelligent agent modules are highly decoupled, supporting cross-department migration and deployment, and can be extended to other specialty fields by adjusting the auditing points and knowledge base.

[0063] ​​In the implementation process, the system first calls the hybrid retrieval and rearrangement module, combines the BM25 keyword retrieval and vector semantic retrieval mechanisms, and recalls the literature and evidence-based data most relevant to the current case characteristics from the knowledge base. Then, the cross-encoder is used to score the relevance of the candidate literature, and the Z-score standardization and linear fusion strategy are used to optimize the ranking, and finally the Top-K evidence-based context is selected.

[0064] The expert diagnosis agent generates diagnosis basis, preliminary diagnosis, differential diagnosis and treatment plan accordingly, and attaches reference clues and literature sources to enhance the traceability and professional reliability of the suggestion content. Finally, the system splices the case characteristics with the generated diagnosis basis, preliminary diagnosis, differential diagnosis and treatment plan to form a complete first course record, providing a structured basis for subsequent clinical review and evidence-based recommendation.

[0065] Through the above steps, the present application realizes the intelligent, structured and closed-loop management of the whole process of pre-consultation in urology surgery. The system uses the multi-agent collaboration mechanism to organically link the consultation, drafting, review, confirmation and diagnosis suggestion generation links to form a dynamic interaction and automatic correction workflow. Each agent has clear division of labor and collaborative operation under the guidance of the node-routing graph, ensuring complete information collection, consistent logic and efficient response. By introducing the dialogue state object, the system realizes the unified management and dynamic update of the consultation context, so that each information interaction can be traced. With the hybrid retrieval, semantic matching and evidence-based reasoning mechanism, the diagnosis and treatment plan generated by the system has both scientificity and individualization. The system not only improves the work efficiency of doctors and the participation of patients, but also has important application value and promotion significance in medical data standardization, knowledge sedimentation and auxiliary decision-making.

[0066] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above figures and description are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A pre-diagnosis method for urology based on multi-agent collaboration, comprising the following steps: 1) Dialogue information collection: The system interacts with the patient in natural language through a dialogue collection agent, automatically guiding the collection of the patient's basic information, chief complaint symptoms, accompanying symptoms, etc., and recording them in a structured manner in the dialogue state object; 2) Case Feature Compilation: Based on the dialogue information, the expert compilation AI extracts and fills in case feature fields, including present medical history, past medical history, and physical examination, and generates subsequent consultation control instructions to drive the system to continue collecting supplementary information; 3) Completeness review: The completeness review agent checks the structural and content completeness of the case based on the key points of urology. If it is incomplete, it returns a minimized follow-up instruction, forming a local closed-loop mechanism of data collection-drafting-review. 4) Patient Confirmation and Arbitration: The system displays the preliminary draft to the patient, and the arbitration agent identifies the type of patient feedback. If there are revision suggestions, the drafting and review process is iterated again to ensure patient participation and the accuracy of medical records. 5) Initial progress note generation: The expert diagnostic agent calls the hybrid retrieval and rearrangement module to complete literature retrieval and fine-tuning based on case characteristics, generate diagnostic basis, preliminary diagnosis, differential diagnosis and treatment plan with evidence-based citation, and output a complete and structured initial progress note.

2. The urological pre-diagnosis method based on multi-agent collaboration as described in claim 1, characterized in that: A node-routing graph mechanism is used to clarify the state transition paths between agents, enabling collaboration and decoupling among modules; The dialogue state object is used for context recording and updating throughout the entire consultation process; The hybrid retrieval and rearrangement module combines BM25 retrieval with vector semantic matching, supplemented by a cross encoder and Z-score normalization method, to recommend evidence-based literature.

3. A urological pre-diagnosis system based on multi-agent collaboration, characterized in that, Multi-agent collaboration is driven by a node-routing graph, where each module acts as an independent agent node and its state transitions are controlled by predefined routing conditions. Dialogue state objects are used throughout the entire system process to uniformly record, transmit, and update interactive information and control commands; include: The dialogue information collection module is used to collect patients' basic information, chief complaints, accompanying symptoms, etc. through natural language interaction, and store the information in a structured manner in the dialogue state object; The case characteristics drafting module is used to extract and fill in the present medical history, past medical history, physical examination and other contents based on the collected information, and generate consultation control instructions; The integrity verification module is used to automatically detect the structure and information integrity of medical records based on the key points of urology. If it is incomplete, it generates a minimal follow-up question instruction and returns it to the data collection process. The Patient Confirmation and Arbitration module is used to present the initial draft to patients and identify feedback types, respond to confirmations, revisions, or questions, and drive the revision process. The initial medical record generation module calls the hybrid search and rearrangement module to retrieve and screen literature based on case characteristics, and generates diagnostic basis, preliminary diagnosis, differential diagnosis and treatment plan containing evidence-based citations.

4. The system according to claim 3, wherein the node-routing graph defines the transfer paths and triggering conditions between agents, realizing closed-loop control and collaborative decoupling of the process.

5. The system according to claim 3, characterized in that, The dialogue state object includes the following fields: message record, number of consultation rounds, consultation control instructions, case characteristics, diagnostic basis, preliminary diagnosis, differential diagnosis, treatment plan and specific revision points.

6. The system according to claim 3, characterized in that, The hybrid retrieval and reordering module combines BM25 keyword retrieval with vector semantic matching, supplemented by cross-encoder scoring and Z-score normalization, to achieve high-quality, evidence-based literature ranking and recommendation.

7. The system according to claim 3, characterized in that, The integrity review module uses a pre-set list of key points in urology to guide the system in collecting missing information through a minimal follow-up questioning approach, thereby achieving a multi-round iterative completion mechanism.

8. The system according to claim 3, characterized in that, The arbitration module uses a semantic recognition model to determine the type of patient feedback, automatically extracts key revision points, and drives the process back to the final draft and review process.

9. The system according to claim 3, characterized in that, The system is adapted to the structured format of electronic medical records, and the generated content supports subsequent clinical decision-making, teaching retrospection, and the construction of medical knowledge bases.

10. The system according to claim 3, characterized in that, The various intelligent agent modules are highly decoupled, supporting cross-departmental migration and deployment. By adjusting the review criteria and knowledge base, it can be extended to other specialized fields.

Images