Medical diagnosis reasoning method and system based on multi-agent coevolution

Abstract

The invention provides a medical diagnosis reasoning method and system based on multi-agent coevolution, and the method comprises the steps: driving a large-scale language model to decompose a medical problem through a problem decomposition agent, and determining a to-be-retrieved sub-problem through a self-adaptive retrieval agent, and selecting a proper knowledge base and a proper retrieval algorithm to obtain multi-source retrieval data, and verifying that the intelligent agent performs de-reforming synthesis on the data to obtain an integrated abstract. The expert recruitment agent builds an expert team according to the abstract, the expert agent diagnoses and reasones to build a dynamic reasoning library, the reasoning library is iteratively optimized through judgment, cross reasoning and the auto-reflection physician agent, the dynamic convergence controller monitors the reasoning library, iteration is terminated when conditions are met, and a final diagnosis decision is output. According to the method, information is directionally obtained from a plurality of authoritative and professional medical knowledge bases through problem multi-granularity retrieval and population evolutionary reasoning, and reasoning efficiency can be improved while reasoning correctness is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical decision-making driven by artificial intelligence, and particularly relates to a medical diagnosis reasoning method and system based on multi-agent collaborative evolution. BACKGROUND

[0002] Multi-Agent Systems (MAS) as a kind of distributed artificial intelligence technology, in recent years, has obtained extensive attention and application in the field of medical health. This kind of system can effectively process complex diagnosis reasoning, treatment plan making and other tasks in the medical field through the collaborative work of multiple intelligent agents with autonomous decision-making ability. In the medical diagnosis scene, the MAS system is usually composed of multiple specialized agents, each of which focuses on the knowledge processing of a specific medical field, and through the exchange and cooperation of information among intelligent agents, comprehensive analysis and judgment of complex cases are realized.

[0003] Current medical MAS systems are mainly applied in the following important fields: first, disease diagnosis support system, this kind of system integrates clinical symptoms, laboratory examination results and medical images and other multi-modal data to provide auxiliary diagnosis suggestions for doctors; second, treatment plan recommendation system, based on the individual characteristics of patients and the latest clinical guidelines, personalized treatment suggestions are generated; in addition, there are also medical resource scheduling system, remote medical collaboration platform, etc. The common feature of these application scenarios is that they all need to process a large amount of heterogeneous medical data and make complex decisions based on multi-dimensional information.

[0004] Although MAS technology has shown great potential in the medical field, there are still several key technical bottlenecks in actual application, which seriously affect the reliability and practicality of the system. The most prominent problems include the knowledge illusion phenomenon and the low reasoning efficiency.

[0005] The knowledge illusion problem mainly manifests as factual errors or logical contradictions in the medical information generated by the system. This phenomenon is particularly common in MAS systems based on large language models, when the system encounters rare cases or complex symptom combinations that are not fully covered in the training data, the intelligent agent may produce a seemingly reasonable but actually incorrect diagnosis conclusion. More seriously, such errors often have considerable deceptive nature, which may mislead clinical decision-making and pose a potential threat to patient safety.

[0006] The reasoning efficiency problem is reflected in the computational resource consumption and response time of the system when dealing with complex medical cases. Existing MAS systems often require multiple agents to engage in multiple rounds of discussion and argumentation when dealing with multi-system diseases or rare disease differential diagnosis. In this process, a large number of repeated or inefficient interactions between agents may occur, resulting in system response delay and waste of computing resources. In real-time clinical decision-making scenarios, this efficiency problem may directly affect the timeliness of diagnosis and treatment. SUMMARY

[0007] In view of the above situation, the main purpose of the present application is to propose a medical diagnosis reasoning method and system based on multi-agent co-evolution, in order to solve the above technical problems.

[0008] The present application proposes a medical diagnosis reasoning method based on multi-agent co-evolution, which comprises the following steps: Step 1, input the medical problem, use the problem decomposition agent to drive the large language model to decompose the medical problem, and get the sub-content block; Step 2, use the adaptive retrieval agent to drive the large language model to process the sub-content block, and get the sub-problem set that needs to be retrieved; based on the sub-problem set that needs to be retrieved, use the sub-problem selection knowledge base and retrieval algorithm to retrieve, to get multi-source retrieval data; Step 3, use the verification agent to drive the large language model to sequentially process the multi-source retrieval data for deduplication and integration, to get the integrated summary; Step 4, use the expert recruitment agent to drive the large language model to instantiate the medical problem and the integrated summary, to get the expert agent team; Step 5, based on the expert agent team, use the expert agent to drive the large language model to diagnose and reason the medical problem, to get the dynamic reasoning library; Step 6, use the large language model to construct the judge doctor agent, cross-reasoning doctor agent and self-reflection doctor agent by using the pre-set role definition prompt word and function definition prompt word respectively; based on the dynamic reasoning library, use the judge doctor agent, cross-reasoning doctor agent and self-reflection doctor agent respectively for adaptive evaluation, advantage cross-fusion and defect correction mutation processing in an iterative manner, to get the continuously optimized dynamic reasoning library; Step 7, based on the continuously optimized dynamic reasoning library, use the dynamic convergence controller to monitor the iteration state, and terminate the iteration when any of the consensus convergence, high-score path and single-path convergence conditions are met, and output the final diagnosis decision.

[0009] The application further provides a medical diagnosis reasoning system based on multi-agent collaborative evolution, wherein the system applies the medical diagnosis reasoning method based on multi-agent collaborative evolution as described above, and the system comprises: A multi-granularity retrieval enhancement generator module: An input medical question is decomposed by a question decomposition agent to drive a large language model to obtain sub-content blocks; The sub-content blocks are processed by an adaptive retrieval agent to drive the large language model to obtain a set of sub-questions that need to be retrieved; based on the set of sub-questions that need to be retrieved, a knowledge base and a retrieval algorithm are selected to retrieve the set of sub-questions that need to be retrieved to obtain multi-source retrieval data; The multi-source retrieval data are sequentially processed by a verification agent to drive the large language model to obtain an integrated summary; A population evolution reasoning module: The medical question and the integrated summary are instantiated by an expert recruitment agent to drive the large language model to obtain an expert agent team; Based on the expert agent team, the medical question is diagnosed and reasoned by the expert agent to drive the large language model to obtain a dynamic reasoning library; The large language model is used to construct a judge doctor agent, a cross-reasoning doctor agent and a self-reflection doctor agent by using pre-set role definition prompt words and function definition prompt words; based on the dynamic reasoning library, the judge doctor agent, the cross-reasoning doctor agent and the self-reflection doctor agent are used for adaptive evaluation, advantage cross-fusion and defect correction mutation processing in an iterative manner to obtain a continuously optimized dynamic reasoning library; A dynamic convergence controller module: Based on the continuously optimized dynamic reasoning library, a dynamic convergence controller is used to monitor the iteration state, and the iteration is terminated when any of the consensus convergence, high-score path and single-path convergence conditions is met, and a final diagnosis decision is output.

[0010] Compared with the prior art, the application has the following beneficial effects: 1. The application fundamentally suppresses the "illusion" phenomenon of the large language model in the professional field, and the core lies in the front-end multi-granularity retrieval enhancement generator module, which can obtain information from multiple authoritative and professional medical knowledge bases through question decomposition and adaptive retrieval, ensuring that all knowledge used for reasoning is traceable, thereby avoiding arbitrary component formation. The verification agent removes and logically integrates the multi-source information retrieved to form a high-quality, non-redundant and internally consistent structured knowledge summary. This provides a solid and reliable "fact base" for subsequent reasoning, and fundamentally eliminates the reasoning illusion caused by contradictory or incorrect information; 2、The application significantly improves the accuracy and reliability of the diagnostic conclusion by simulating multidisciplinary expert consultation and Darwinian survival of the fittest through the population evolution reasoning module. Through accurate population initialization, a team of intelligent agents containing different specialist perspectives can be created to conduct comprehensive and multi-angle examination of complex cases, effectively overcoming the diagnostic limitations that may be caused by a single model or single perspective. By introducing the evolution mechanisms of "dominant crossover" and "directed variation", the optimal reasoning path can be automatically integrated, and the defective path can be reflected and corrected. This continuous iteration and self-improvement process can effectively identify and correct weak links in the reasoning chain, gradually approach and even surpass the accuracy of traditional consultation, and effectively avoid falling into a local optimal solution; 3、The application realizes intelligent start and stop of the reasoning process by introducing a dynamic convergence controller module, greatly improves the running efficiency while ensuring the diagnostic quality, effectively avoids redundant calculation consumption, and can automatically identify whether the reasoning process has converged to a stable optimal solution. This flexible termination strategy not only quickly locks the high-quality diagnostic conclusion, but also avoids the intelligent agent continuing to have ineffective "sub-optimal discussion" after the optimal solution is formed, thereby significantly shortening the diagnosis time, saving the calculation resources, and achieving the best balance between diagnosis efficiency and accuracy.

[0011] Additional aspects and advantages of the application will be made apparent by the following description. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 A step flow chart of a medical diagnosis reasoning method based on multi-agent collaborative evolution is proposed for the application; Figure 2 A method framework diagram of a medical diagnosis reasoning method based on multi-agent collaborative evolution is proposed for the application; Figure 3 A system structure diagram of a medical diagnosis reasoning system based on multi-agent collaborative evolution is proposed for the application. DETAILED DESCRIPTION

[0013] The embodiments of the application are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference signs 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.

[0014] These and other aspects of embodiments of the present application will become clear from the following description and the accompanying drawings. In the description and drawings, particular embodiments of the application are disclosed in connection with specific details to provide a thorough understanding of embodiments of the application. It will be apparent, however, to one skilled in the art that embodiments of the application can be practiced without many of these specific details. In other instances, well-known structures and functions have not been described in detail in order to avoid obscuring the understanding of embodiments of the application.

[0015] Referring to Figure 1 The embodiment provides a medical diagnosis reasoning method based on multi-agent cooperative evolution, and the method comprises the following steps: Step 1, inputting a medical problem, decomposing the medical problem by using a problem decomposition agent to drive a large language model to obtain a sub-content block.

[0016] Referring to Figure 2 In step 1, a medical problem is inputted, the medical problem is decomposed by using a problem decomposition agent to drive a large language model to obtain a sub-content block, and the following relationship exists in the corresponding process: ; Among them, denotes a sub-content block, denotes processing by a deepseek large language model, denotes a medical problem, denotes a prompt word for executing a function of a problem decomposition agent, denotes a role definition of the problem decomposition agent.

[0017] Step 2, processing the sub-content block by using an adaptive retrieval agent to drive a large language model to obtain a sub-problem set that needs to be retrieved; based on the sub-problem set that needs to be retrieved, a knowledge base and a retrieval algorithm are selected by using the sub-problem that needs to be retrieved to perform retrieval, so as to obtain multi-source retrieval data.

[0018] In step 2, the decomposed sub-content block is processed by using an adaptive retrieval agent to drive a large language model to obtain a sub-problem set that needs to be retrieved; based on the sub-problem set that needs to be retrieved, a knowledge base and a retrieval algorithm are selected by using the sub-problem that needs to be retrieved to perform retrieval, so as to obtain multi-source retrieval data, and the following relationship exists in the corresponding process: ; Among them, denotes the th sub-problem that needs to be retrieved, denotes an index of the sub-problem that needs to be retrieved, denotes the th knowledge base matched with the sub-problem that needs to be retrieved, denotes the th retrieval algorithm matched with the sub-problem that needs to be retrieved, prompt words representing functions performed by the adaptive retrieval agent, role definition of the adaptive retrieval agent.

[0019] Step 3: Use the verification agent to drive the large language model to sequentially perform deduplication and integration processing on the multi-source retrieval data to obtain the integrated summary.

[0020] In step 3, the verification agent is used to drive the large language model to sequentially perform deduplication and integration processing on the multi-source retrieval data to obtain the integrated summary. The following relationship exists in the corresponding process: ; Among them, represents the integrated summary, represents the multi-source retrieval data, represents prompt words representing functions performed by the verification agent, represents the role definition of the verification agent.

[0021] Step 4: Use the expert recruitment agent to drive the large language model to instantiate the medical problem and the integrated summary to obtain the expert agent team.

[0022] In step 4, the expert recruitment agent is used to drive the large language model to instantiate the medical problem and the integrated summary to obtain the expert agent team. The following relationship exists in the corresponding process: ; Among them, represents the expert agent team, both represent expert agents, represents the index of the expert agent, represents prompt words representing functions performed by the expert recruitment agent, represents the role definition of the expert recruitment agent.

[0023] Step 5: Based on the expert agent team, use the expert agent to drive the large language model to diagnose and reason the medical problem to obtain the dynamic reasoning library.

[0024] In step 5, based on the expert agent team, the expert agent is used to drive the large language model to diagnose and reason the medical problem to obtain the dynamic reasoning library. The following relationship exists in the corresponding process: ; Among them, represents the path of the expert agent reasoning, represents prompt words representing functions performed by the expert agent, represents the role definition of the first expert agent, represents a dynamic reasoning library.

[0025] Step 6, using a large language model, a pre-set role definition prompt word and a function definition prompt word are used to construct a judge physician agent, a cross-reasoning physician agent and a self-reflection physician agent; based on the dynamic reasoning library, the judge physician agent, the cross-reasoning physician agent and the self-reflection physician agent are used in an iterative form for adaptive evaluation, advantage cross-fusion and defect correction mutation processing to obtain a continuously optimized dynamic reasoning library.

[0026] In step 6, based on the dynamic reasoning library, the judge physician agent, the cross-reasoning physician agent and the self-reflection physician agent are used in an iterative form for adaptive evaluation, advantage cross-fusion and defect correction mutation processing to obtain a continuously optimized dynamic reasoning library, which includes the following sub-steps: S601, based on the dynamic reasoning library, the judge physician agent is used to drive the large language model to perform adaptive evaluation processing on the reasoning path to obtain a path score set; based on the path score set, the reasoning paths in the dynamic reasoning library are sorted in descending order of path score to obtain a sorted reasoning library, and the following relationship exists in the corresponding process: ; Wherein, represents the score of the path reasoned by the first expert agent, represents the prompt word for the function of the judge physician agent, represents the role definition of the judge physician agent; S602, using the cross-reasoning physician agent to drive the large language model to perform cross-fusion processing on the first two reasoning paths in the sorted reasoning library to obtain a cross-reasoning path, and the following relationship exists in the corresponding process: ; Wherein, represents the two reasoning paths that need to be crossed, represents the function of selecting the two reasoning paths with the highest and second highest adaptive scores in the dynamic reasoning library, represents the cross-reasoning path, represents the prompt word for the function of the cross-reasoning physician agent, represents the role definition of the cross-reasoning physician agent; S603, the last reasoning path in the sorted reasoning library is processed by directional variation using the self-reflective physician intelligent agent driving a large language model, and a mutated reasoning path is obtained, and the following relationship exists in the corresponding process: ; Among them, represents the reasoning path that needs to be mutated, represents the function of selecting the reasoning path with the lowest adaptability score in the dynamic reasoning library, represents the mutated reasoning path, represents the prompt word of the self-reflective physician intelligent agent, represents the role definition of the self-reflective physician intelligent agent; S604, the first two reasoning paths and the last reasoning path in the sorted reasoning library are removed to obtain an updated reasoning library; the crossed reasoning path and the mutated reasoning path are added to the updated reasoning library to obtain a first iteration reasoning library. In the form of iteration, the first iteration reasoning library is repeatedly inputted into steps S601 to S604 for a preset number of times to obtain a continuously optimized dynamic reasoning library.

[0027] Step 7, based on the continuously optimized dynamic reasoning library, the iteration state is monitored using a dynamic convergence controller, and the iteration is terminated when any of the consensus convergence, high-score path, and single-path convergence conditions are met, and the final diagnostic decision is output.

[0028] In step 7, based on the continuously optimized dynamic reasoning library, the iteration state is monitored, and the iteration is terminated when any of the consensus convergence, high-score path, and single-path convergence conditions are met, and the final diagnostic decision is output, and the following relationship exists in the corresponding process: ; Among them, represents that there is a reasoning path , represents that the following conditions are met, represents the number of reasoning paths in the dynamic reasoning library, represents the number of reasoning paths with the same result as in the dynamic reasoning library, represents the score of reasoning path in the th iteration, represents the score of reasoning path ; represents the number of reasoning paths with the same result as in the dynamic reasoning library.

[0029] It should be noted that in the process of iteration, the iteration state is monitored by a dynamic convergence controller until any of the following preset conditions is met, and then the iteration is terminated: 1. When most expert agents reach a consensus on a diagnosis result through voting, the iteration will be terminated in advance, and the consensus will be taken as the final diagnosis decision; 2. If there is a reasoning path with an adaptive score higher than 90 in the iteration process, it means that a high-quality diagnostic path has been successfully generated, and the result will be immediately adopted as the final diagnosis decision; 3. If none of the above conditions is met, the iteration will continue until a single optimal solution is converged after multiple iterations, which means that the team of expert agents has found an optimized and highly consistent final diagnosis solution through co-evolution. At this time, the iteration will be terminated and the record will be output as the final diagnosis decision; This dynamic convergence strategy ensures that the quality of diagnosis is guaranteed while effectively controlling the consumption of computing resources and time, achieving reliable and efficient diagnostic conclusions.

[0030] Please refer to Figure 3 The embodiment also provides a medical diagnosis reasoning system based on multi-agent co-evolution, wherein the system applies the medical diagnosis reasoning method based on multi-agent co-evolution as described above, and the system comprises: A multi-granularity retrieval enhancement generator module: Input the medical question, and use the question decomposition agent to drive the large language model to decompose the medical question and obtain sub-content blocks; Use the adaptive retrieval agent to drive the large language model to process the sub-content blocks and obtain a set of sub-questions that need to be retrieved; based on the set of sub-questions that need to be retrieved, use the sub-questions that need to be retrieved to select the knowledge base and retrieval algorithm for retrieval to obtain multi-source retrieval data; Use the verification agent to drive the large language model to sequentially process the multi-source retrieval data for deduplication and integration to obtain an integrated summary; A population evolution reasoning module: Use the expert recruitment agent to drive the large language model to instantiate the medical question and the integrated summary to obtain a team of expert agents; Based on the team of expert agents, use the expert agent to drive the large language model to perform diagnostic reasoning on the medical question to obtain a dynamic reasoning library; The large language model is used to construct a judge physician agent, a cross-reasoning physician agent and a self-reflection physician agent by using pre-set role definition prompts and function definition prompts; and based on a dynamic reasoning library, the judge physician agent, the cross-reasoning physician agent and the self-reflection physician agent are used in an iterative form to perform adaptive evaluation, advantage cross-fusion and defect correction and variation processing, so as to obtain a continuously optimized dynamic reasoning library. A dynamic convergence controller module: Based on the continuously optimized dynamic reasoning library, a dynamic convergence controller is used to monitor an iteration state, and iteration is terminated when any of consensus convergence, high-score path and single-path convergence conditions are met, and a final diagnosis decision is output.

[0031] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0032] It should be understood that each part of the present application can be realized by hardware, software, firmware or a combination thereof. In the above-described embodiments, multiple steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized by hardware, as in another embodiment, it can be realized by any one or a combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0033] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0034] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A medical diagnosis reasoning method based on multi-agent co-evolution, characterized in that, The method comprises the following steps: Step 1, inputting a medical problem, decomposing the medical problem by using a problem decomposition intelligent agent to drive a large language model to obtain a sub-content block; Step 2, processing the sub-content block by using an adaptive retrieval intelligent agent to drive the large language model to obtain a sub-problem set requiring retrieval; based on the sub-problem set requiring retrieval, selecting a knowledge base and a retrieval algorithm for retrieval by using the sub-problem requiring retrieval to obtain multi-source retrieval data; Step 3, sequentially performing deduplication and integration processing on the multi-source retrieval data by using a verification intelligent agent to drive the large language model to obtain an integrated summary; Step 4, instantiating the medical problem and the integrated summary by using an expert recruitment intelligent agent to drive the large language model to obtain an expert intelligent agent team; Step 5, based on the expert intelligent agent team, performing diagnostic reasoning on the medical problem by using an expert intelligent agent to drive the large language model to obtain a dynamic reasoning library; Step 6, using the large language model to construct a judge physician intelligent agent, a cross-reasoning physician intelligent agent and a self-reflection physician intelligent agent by using pre-set role definition prompt words and function definition prompt words; based on the dynamic reasoning library, iteratively performing adaptive evaluation, advantage cross-fusion and defect correction mutation processing on the judge physician intelligent agent, the cross-reasoning physician intelligent agent and the self-reflection physician intelligent agent to obtain a continuously optimized dynamic reasoning library; Step 7, based on the continuously optimized dynamic reasoning library, monitoring the iteration state by using a dynamic convergence controller, and terminating the iteration when any of the consensus convergence, high-score path and single-path convergence conditions is met, and outputting a final diagnosis decision. 2.The multi-agent co-evolution based medical diagnosis reasoning method according to claim 1, wherein, In the step 1, the medical problem is inputted, and the medical problem is decomposed by using a problem decomposition intelligent agent to drive a large language model to obtain a sub-content block, and the following relationship exists in the corresponding process: ； wherein, represents a sub-content block, represents processing by a deepseek large language model, represents a medical question, represents a prompt word for a question decomposition agent to perform a function, represents a role definition for a question decomposition agent. 3.The multi-agent co-evolution based medical diagnosis reasoning method according to claim 2, wherein, In the step 2, the sub-content block is processed by using an adaptive retrieval intelligent agent to drive the large language model to obtain a sub-problem set requiring retrieval; based on the sub-problem set requiring retrieval, a knowledge base and a retrieval algorithm are selected for retrieval by using the sub-problem requiring retrieval to obtain multi-source retrieval data, and the following relationship exists in the corresponding process: ； in, Indicates the first One sub-question that needs to be retrieved. This represents the index of the sub-question that needs to be retrieved. Indicates the first The knowledge base that matches the sub-question that needs to be retrieved. Indicates the first The retrieval algorithm that matches the sub-problems that need to be retrieved. Hint words indicating the functions performed by the adaptive retrieval agent. This represents the role definition of the adaptive retrieval agent.

4. The multi-agent co-evolution based medical diagnosis reasoning method according to claim 3, characterized in that, In the step 3, the multi-source retrieval data are sequentially subjected to deduplication and integration processing by using a verification intelligent agent to drive the large language model to obtain an integrated summary, and the following relationship exists in the corresponding process: ； wherein, represents an integrated summary, represents multi-source search data, represents a prompt word for verifying the function of the intelligent agent, represents a role definition of the verification intelligent agent.

5. The multi-agent co-evolution based medical diagnosis reasoning method according to claim 4, wherein, In the step 4, the medical problem and the integrated summary are instantiated by using an expert recruitment intelligent agent to drive the large language model to obtain an expert intelligent agent team, and the following relationship exists in the corresponding process: ； wherein, represents a team of expert agents, each represents an expert agent, represents an index of the expert agent, represents a prompt word for the expert recruiting agent to perform a function, represents a role definition of the expert recruiting agent.

6. The multi-agent co-evolution based medical diagnosis reasoning method according to claim 5, wherein, In the step 5, based on the expert intelligent agent team, the medical problem is subjected to diagnostic reasoning by using an expert intelligent agent to drive the large language model to obtain a dynamic reasoning library, and the following relationship exists in the corresponding process: ； wherein, represents a path of reasoning for the expert agent, represents a prompt word for the expert agent to perform a function, represents a role definition for the expert agent, represents a dynamic reasoning library.

7. The multi-agent co-evolution based medical diagnosis reasoning method according to claim 6, characterized in that, In the step 6, based on the dynamic reasoning library, iteratively performing adaptive evaluation, advantage cross-fusion and defect correction mutation processing on the judge physician intelligent agent, the cross-reasoning physician intelligent agent and the self-reflection physician intelligent agent to obtain a continuously optimized dynamic reasoning library, which specifically comprises the following sub-steps: S601, based on the dynamic reasoning library, the reasoning path is adaptively evaluated by driving the large language model with the judge doctor agent, and a path score set is obtained; based on the path score set, the reasoning paths in the dynamic reasoning library are sorted in descending order of path score, and a sorted reasoning library is obtained; S602, the first two reasoning paths in the sorted reasoning library are cross-fused by driving the large language model with the cross-reasoning doctor agent, and a cross-reasoning path is obtained; S603, the last reasoning path in the sorted reasoning library is directionally mutated by driving the large language model with the self-reflection doctor agent, and a mutated reasoning path is obtained; S604, the first two reasoning paths and the last reasoning path in the sorted reasoning library are removed to obtain an updated reasoning library; the cross-reasoning path and the mutated reasoning path are added to the updated reasoning library to obtain a first iteration reasoning library; The first iteration reasoning library is repeatedly inputted into steps S601 to S604 for a preset number of times in an iterative manner to obtain a continuously optimized dynamic reasoning library. 8.The multi-agent co-evolution based medical diagnosis reasoning method according to claim 7, wherein, In the step of adaptively evaluating the reasoning path by driving the large language model with the judge doctor agent based on the dynamic reasoning library, a path score set is obtained; in the step of sorting the reasoning paths in the dynamic reasoning library in descending order of path score based on the path score set, a sorted reasoning library is obtained, and the following relationship exists in the corresponding process: ； wherein, represents a score of a path of reasoning of the expert agent, represents a prompt word for the referee physician agent to perform a function, represents a role definition of the referee physician agent; In the step of cross-fusing the first two reasoning paths in the sorted reasoning library by driving the large language model with the cross-reasoning doctor agent, the following relationship exists in the corresponding process: ； wherein, represents two reasoning paths that need to be crossed, represents a function function of selecting two reasoning paths with the highest and second highest adaptive scores in the dynamic reasoning library, represents the reasoning path after crossing, represents a prompt word for the cross-reasoning physician agent to execute functions, represents the role definition of the cross-reasoning physician agent; In the step of directionally mutating the last reasoning path in the sorted reasoning library by driving the large language model with the self-reflection doctor agent, the following relationship exists in the corresponding process: ； wherein, represents a reasoning path that needs to be varied, represents a function function of selecting a reasoning path with the lowest adaptability score in the dynamic reasoning library, represents a varied reasoning path, represents a prompt word of the self-reflective physician agent executing a function, represents a role definition of the self-reflective physician agent. 9.The multi-agent co-evolution based medical diagnosis reasoning method according to claim 8, wherein, In step 7, based on the continuously optimized dynamic reasoning library, the iteration state is monitored, and when any of the consensus convergence, high-score path, and single-path convergence conditions is met, the iteration is terminated, and the final diagnosis decision is output, and the following relationship exists in the corresponding process: ； in, This indicates the existence of a reasoning path. , This indicates that the following conditions are met. This indicates the number of inference paths in the dynamic inference library. This indicates that the calculation in the dynamic inference library is related to... The number of results with the same reasoning path. Indicates the first Inference path in round iteration The rating, Indicating the reasoning path The rating; Indicates the first The number of identical inference paths in the inference library during each iteration. 10.A medical diagnosis reasoning system based on multi-agent co-evolution, characterized in that, The system applies the medical diagnosis reasoning method based on multi-agent co-evolution of any one of claims 1 to 9, and the system comprises: A multi-granularity retrieval enhancement generator module: Input a medical question, decompose the medical question by driving a large language model with a question decomposition agent to obtain a sub-content block; Process the sub-content block by driving a large language model with an adaptive retrieval agent to obtain a set of sub-questions that need to be retrieved; based on the set of sub-questions that need to be retrieved, select a knowledge base and a retrieval algorithm for retrieval to obtain multi-source retrieval data; Integrate the multi-source retrieval data by driving a large language model with a verification agent to obtain an integrated summary; A population evolution reasoning module: Instantiates the medical question and the integrated summary by driving a large language model with an expert recruitment agent to obtain an expert agent team; Based on the team of expert agents, a large language model driven by expert agents is used to diagnose and reason medical problems to obtain a dynamic reasoning library; Using a large language model, a pre-defined role definition prompt word and a function definition prompt word are used to construct a judge physician agent, a cross-reasoning physician agent, and a self-reflection physician agent; Based on the dynamic reasoning library, the judge physician agent, the cross-reasoning physician agent, and the self-reflection physician agent are used for adaptive evaluation, advantage cross-fusion, and defect correction and mutation processing in an iterative form to obtain a continuously optimized dynamic reasoning library. Dynamic convergence controller module: Based on the continuously optimized dynamic reasoning library, a dynamic convergence controller is used to monitor the iteration state, and when any of the consensus convergence, high-score path, and single-path convergence conditions are met, the iteration is terminated, and the final diagnosis decision is output.

Images