Self-evolution medical consultation system based on multiple agents and large language model

Abstract

The invention provides a self-evolution medical consultation system based on multiple agents and a large language model, and the system comprises an obtaining module which is used for obtaining a target information set from a shared dialogue sample pool, the target information set comprises M sample information combinations, and the sample information combinations comprise N rounds of inquiry information, n rounds of inquiry information included in each sample information combination are inquiry information of the same agent and the target patient, the agents corresponding to different sample information combinations are different, and M and N are integers greater than 1; and the output module is used for outputting diagnosis information of the target patient according to the target information set. Therefore, the diagnosis information output efficiency of the target patient and the accuracy of the output result are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to a self-evolving medical consultation system based on multi-agents and a large language model. Background Art

[0002] With the continuous development of artificial intelligence technology, the functions of large language models are becoming more and more abundant. Currently, large language models can be widely used in medical consultation. However, when generating diagnostic suggestions, large language models tend to generate diagnostic information with low accuracy. It can be seen that the current diagnostic effect of large language models is poor. Summary of the invention

[0003] The embodiment of the present application provides a self-evolving medical consultation system based on multi-agent and large language model to solve the problem of poor diagnostic effect of current large language models.

[0004] To solve the above problems, this application is implemented as follows:

[0005] In a first aspect, the present application embodiment provides a self-evolving medical consultation system based on multi-agent and large language model, including:

[0006] An acquisition module is used to acquire a target information set from a shared conversation sample pool, wherein the target information set includes M sample information combinations, wherein the sample information combination includes N rounds of consultation information, wherein the N rounds of consultation information included in each sample information combination are consultation information between the same agent and the target patient, and different sample information combinations correspond to different agents, and both M and N are integers greater than 1;

[0007] An output module is used to output the diagnostic information of the target patient according to the target information set.

[0008] In a second aspect, the present application embodiment provides a medical consultation method, comprising:

[0009] Obtain a target information set from the shared conversation sample pool, wherein the target information set includes M sample information combinations, and the sample information combination includes N rounds of consultation information. The N rounds of consultation information included in each sample information combination are consultation information between the same agent and the target patient, and different sample information combinations correspond to different agents. M and N are both integers greater than 1.

[0010] The diagnostic information of the target patient is output according to the target information set.

[0011] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps in the method described in the second aspect above.

[0012] In a fourth aspect, an embodiment of the present application further provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method described in the second aspect above.

[0013] In a fifth aspect, an embodiment of the present application further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps in the method described in the second aspect above.

[0014] In an embodiment of the present application, a target information set is obtained from a shared conversation sample pool, the target information set including M sample information combinations, the sample information combination including N rounds of consultation information, the N rounds of consultation information included in each sample information combination are consultation information between the same intelligent agent and the target patient, and different sample information combinations correspond to different intelligent agents, and M and N are both integers greater than 1; the diagnosis information of the target patient is output according to the target information set.

[0015] In this way, since the target information set includes M sample information combinations, and the sample information combinations include N rounds of medical consultation information, by accessing the N rounds of medical consultation information in the target information set and outputting the target patient's diagnostic information based on the N rounds of medical consultation information in the target information set, information redundancy is reduced, and the output efficiency of the target patient's diagnostic information and the accuracy of the output results are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0017] Figure 1 It is a structural diagram of a self-evolving medical consultation system based on multi-agents and a large language model provided in an embodiment of the present application;

[0018] Figure 2 This is an application scenario diagram of a self-evolving medical consultation system based on multi-agents and a large language model provided in an embodiment of the present application;

[0019] Figure 3 is a flow chart of the medical consultation method provided in an embodiment of the present application;

[0020] Figure 4 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] The terms "first", "second" etc. in the embodiments of the present application are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. In addition, the terms "include" and "have" and any variation thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "and / or" is used in the present application to represent at least one of connected objects, such as A and / or B and / or C, representing the inclusion of separate A, separate B, separate C, and A and B all exist, B and C all exist, A and C all exist, and 7 situations that A, B and C all exist.

[0023] See also Figure 1 , Figure 1 Schematic diagram of the structure of a self-evolving medical consultation system based on multi-agent and large language model provided in the embodiment of the present application. Figure 1 As shown, the self-evolving medical consultation system based on multi-agent and large language model may include:

[0024] The acquisition module 101 is used to acquire a target information set from a shared conversation sample pool, wherein the target information set includes M sample information combinations, and the sample information combination includes N rounds of consultation information. The N rounds of consultation information included in each sample information combination are consultation information between the same agent and the target patient, and different sample information combinations correspond to different agents. M and N are both integers greater than 1.

[0025] Among them, the shared dialogue sample pool can also be called a shared dialogue pool, and each intelligent agent corresponding to the M sample information combinations can access the above-mentioned shared dialogue sample pool, and can also store the consultation information between each intelligent agent and the target patient in the shared dialogue sample pool.

[0026] The role of the agent is not specifically limited here. Optionally, the role of the agent may include at least one of the following: general practitioner, internist, surgeon, pediatrician, obstetrician, gynecologist, radiologist, pathologist, pharmacist, etc. For example: Figure 2 , Figure 2A, B, C, D, X, B / D can represent different agent roles, and the final diagnosis information of the target patient can be stored in Figure 2 The database shown in the storage icon (the database can be the correct consultation experience database and chain thinking experience database in the following text).

[0027] The above-mentioned general practitioners, internists, surgeons, pediatricians, and obstetricians and gynecologists can be referred to as attending physician agents or expert physician agents, while radiologists, pathologists, and pharmacists can be referred to as other agents.

[0028] It should be noted that the attending physician agent can be determined based on information such as the target patient's chief complaint, and the roles of other agents can be assigned by the attending physician agent based on information such as the target patient's medical history and medical problems. For example, when the target patient is a child, the attending physician agent can be a pediatrician; when the target patient's medical history includes that the target patient mainly suffers from liver disease, the attending physician agent can be an internist.

[0029] In addition, optionally, when the attending physician agent assigns roles to other agents based on information such as the target patient's medical history and medical problems, a large language model (LLM)-based role assignment can be used to ensure that the selected agent role can accurately handle the target patient's current condition.

[0030] For example, the agent can be determined according to the following steps.

[0031] 1.1. The target patient enters the consultation system with personal background and medical problems, and the consultation system includes multiple agents.

[0032] 1.2. The attending physician agent assigns expert doctor roles (i.e. agent roles) according to the patient's condition. Mandatory agent roles include radiologists, pathologists, and pharmacists.

[0033] 1.3. The attending physician agent uses an LLM-based role allocation model to ensure that the selected role can accurately handle the current condition.

[0034] The information of the number of rounds of consultation of the consultation information included in each sample information combination may be the same, that is, may be N rounds. The value of N is not specifically limited here, and optionally, N may be 2.

[0035] It should be noted that each agent can inquire according to the condition of the target patient in each round of consultation, and can generate suggestion information in each round of consultation, that is, each round of consultation information can include suggestion information, and in the next round of consultation, the suggestion information can be adjusted based on the previous round of consultation information. The above-mentioned consultation method can be called the residual (ResNet-Like) discussion mode. When N is 2, that is, when outputting the diagnosis information of the target patient in each round, only the consultation information of all agents in the previous two rounds can be accessed. In this way, information redundancy and over-reliance on the analysis of other agents are avoided, thereby reducing information pollution, helping agents to maintain more independence in decision-making, preventing collective hallucinations, and improving the accuracy of the output diagnosis information.

[0036] It should be noted that the above-mentioned consultation method can also adopt other forms of discussion modes, such as: DenseNet structure or HRNet structure.

[0037] For example, the consultation can be conducted according to the following steps.

[0038] 2.1. In the first round of consultation, the expert doctor agent independently provides condition analysis and treatment suggestions, which are stored in the shared speech pool (S1).

[0039] 2.2. Starting from the second round of consultation, each agent can access the discussion results of the previous round and make new analysis and suggestions based on them. The results are stored in (S 2,k ).

[0040] 2.3. Starting from the (i+1)th round of consultation, the agent can access the discussion results of the first two rounds (S i ) and (S i-1 ), further refining the recommendations until all physicians reached a consensus.

[0041] 2.4. If consensus is reached, the discussion ends; if no consensus is reached, a voting decision-making mechanism is adopted, and the agent's suggestion is randomly selected when necessary.

[0042] The output module 102 is used to output the diagnosis information of the target patient according to the target information set.

[0043] In an embodiment of the present application, since the target information set includes M sample information combinations, and the sample information combinations include N rounds of medical consultation information, by accessing the N rounds of medical consultation information in the target information set and outputting the target patient's diagnostic information based on the N rounds of medical consultation information in the target information set, information redundancy is reduced, and the output efficiency of the target patient's diagnostic information and the accuracy of the output results are improved.

[0044] It should be noted that the specific method of outputting the diagnostic information of the target patient according to the target information set is not limited here.

[0045] As an optional implementation manner, each round of medical consultation information in the N rounds of medical consultation information includes predictive diagnosis information, and the output module is further used to:

[0046] Acquire a target information from each sample information combination to obtain M target information, wherein the target information is the maximum predicted diagnosis information at the consultation time among the N rounds of consultation information included in each sample information combination;

[0047] In the case that all the M target information are matched, any one of the M target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output.

[0048] Among them, the target information is the predicted diagnosis information with the largest consultation time among the N rounds of consultation information included in each sample information combination, which can be understood as: since the consultation times corresponding to different consultation rounds are different, and the consultation time is proportional to the number of consultation rounds, that is, the larger the number of consultation rounds, the larger the corresponding consultation time. Therefore, the target information is the predicted diagnosis information with the largest consultation time among the N rounds of consultation information included in each sample information combination, which can be understood as the predicted diagnosis information corresponding to the last round of the N rounds of consultation information, and the predicted diagnosis information can refer to the diagnosis information or suggestion information for the target patient.

[0049] In the implementation manner of the present application, when all M target information are matched, it means that different intelligent agents have a high degree of match for the predicted diagnostic information of the target patient and a small amount of diagnostic analysis. Therefore, any one of the M target information can be determined as the diagnostic information of the target patient, thereby improving the output efficiency and accuracy of the diagnostic information of the target patient.

[0050] As an optional implementation manner, the output module is further used for:

[0051] In the case that there are L target information mismatches among the M target information, obtaining the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination, where L is an integer greater than or equal to 1, and L is less than M;

[0052] When the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination is greater than the preset number of rounds, determine the size of L and ML;

[0053] When L is greater than ML, any one of the L target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output;

[0054] In the case where L is less than ML, determining any one of the target information among the M target information except the L target information as the diagnosis information of the target patient, and outputting the diagnosis information of the target patient;

[0055] When L is equal to ML, any one of the M target information except the L target information, or any one of the L target information, or a randomly selected one of the M target information is determined as the diagnostic information of the target patient, and the diagnostic information of the target patient is output.

[0056] The specific value of the preset number of rounds is not limited here. Optionally, the value of the preset number of rounds can be 10.

[0057] In the implementation manner of the present application, when there are L target information mismatches among the M target information, and the information on the number of consultation rounds is greater than the preset number of rounds, it means that after multiple rounds of consultation, the predicted diagnosis information (i.e., target information) output by different intelligent agents still have differences, that is, they have failed to converge at this time. At this time, a voting decision-making mechanism can be used to determine and output the diagnosis information of the target patient to avoid the waste of computing resources caused by an infinite number of consultation rounds and the occurrence of low efficiency in outputting the diagnosis information of the target patient.

[0058] The above voting decision mechanism can be:

[0059] When L is greater than ML, it indicates that the unmatched target information is more than the matched target information, and thus any one of the L unmatched target information can be determined as the diagnostic information of the target patient;

[0060] When L is less than ML, it indicates that the matched target information is more than the unmatched target information, and thus any one of the ML unmatched target information can be determined as the diagnostic information of the target patient;

[0061] When L is equal to ML, it indicates that the matched target information is equal to the unmatched target information, and thus any one of the M target information except the L target information, or any one of the L target information, or a randomly selected one of the M target information can be determined as the diagnostic information of the target patient.

[0062] As an optional implementation manner, the inquiry information included in the N rounds of medical inquiry information is:

[0063] The query information in the sample case information obtained by the corresponding intelligent agent from the preset correct consultation experience database; and / or,

[0064] Inquiry information from sample case information obtained from the preset chain thinking experience database.

[0065] Among them, the correct consulting experience database can be called the correct consulting experience base (CorrectKB), and the chain thinking experience database can be called the chain thinking experience base (ChainKB).

[0066] Among them, the correct consultation experience database can store sample cases that have been successfully consulted before the target patient is consulted, and the chain thinking experience database can store sample cases that have been incorrectly consulted before the target patient, but the reasons for the incorrect consultation have been abstracted. In this way, when the target patient is consulted based on the inquiry information in the sample case information obtained from the chain thinking experience database, the mistakes made in the previous sample cases can be avoided, thereby improving the accuracy of the consultation.

[0067] The inquiry information included in the N rounds of medical consultation information may refer to the questions asked by the intelligent agent to the target patient in each round of medical consultation.

[0068] In the implementation mode of the present application, the inquiry information included in the N rounds of medical consultation information is: the inquiry information in the sample case information obtained by the corresponding intelligent agent from the preset correct consultation experience database; and / or, the inquiry information in the sample case information obtained from the preset chain thinking experience database. In this way, the intelligent agent can inquire the target patient based on the inquiry information in the sample case information in the correct consultation experience database and the chain thinking experience database, avoiding the mistakes made in the previous inquiry of the sample case information, thereby improving the accuracy of inquiries to the target patient.

[0069] As an optional implementation, it also includes:

[0070] A sending module, used for sending the diagnosis information of the target patient to the first agent;

[0071] Wherein, the first agent is used to review the diagnostic information of the target patient, and if it is determined that the diagnostic information of the target patient is correct, store the diagnostic information of the target patient in the correct consultation experience database, and if it is determined that the diagnostic information of the target patient is wrong, send the diagnostic information of the target patient to the second agent;

[0072] The second intelligent agent is used to abstract and summarize the diagnostic information of the target patient, and store the target information obtained after the abstract summary in the chain thinking experience database. The target information includes error information existing in the diagnostic information of the target patient and suggestion information for modifying the error information.

[0073] Among them, the first agent can refer to the safety and ethics reviewer agent, and the second agent can refer to the chain thinking reviewer agent.

[0074] It should be noted that when the first intelligent agent reviews the diagnostic information of the target patient, in addition to determining whether the diagnostic information of the target patient is correct, it can also review whether there are potential security issues, such as whether the diagnostic information of the target patient violates public order and good morals, etc., to ensure the security and rationality of the diagnostic information.

[0075] In the implementation mode of the present application, when the diagnostic information of the target patient is correct, the first intelligent agent can directly store the diagnostic information of the target patient into the correct consultation experience database; when the diagnostic information of the target patient is incorrect, the second intelligent agent can abstractly summarize the diagnostic information of the target patient, and store the target information obtained after the abstract summary into the chain thinking experience database, so that when the next patient comes, similar sample cases can be screened from the correct consultation experience database and the chain thinking experience database based on the background and condition of the next patient, and the next patient can be inquired based on the similar sample cases, thereby improving the efficiency and accuracy of the inquiry to the next patient.

[0076] It should be noted that after each output of diagnostic information, the diagnostic information can be stored in the correct consultation experience database or the chain thinking experience database. In this way, new cases can be continuously added to enrich the knowledge base, and these cases can be used to optimize the decision prompts of the doctor's agent in the next consultation. The size of the experience database increases with the number of consultations, gradually improving the diagnostic accuracy and robustness of the system; at the same time, when the diagnostic information is wrong, it is necessary to abstract and summarize the diagnostic information, and store the target information obtained after the abstract summary in the chain thinking experience database, that is, self-reflection will be carried out, and new diagnostic ideas will be summarized based on past wrong cases to help the agent avoid similar problems in future consultations.

[0077] It should be noted that when storing sample cases in the correct consultation experience database and the chain thinking experience database, the sample cases can be stored in the form of vectors. Therefore, the correct consultation experience database and the chain thinking experience database can also be called vector databases. In this way, when similar sample cases are screened from the correct consultation experience database and the chain thinking experience database based on the background and condition of the next patient, the vectors of the top-k cases most similar to the vector data of the next patient can be screened in the form of cosine similarity, and the above top-k cases can be used as sample cases.

[0078] For example, you can refer to the following steps to store diagnostic information.

[0079] 3.1. Once the consultation results (i.e., diagnostic information) of the target patient are generated, they will be reviewed by the safety and ethics reviewer agent to identify potential safety issues.

[0080] 3.2. If the diagnostic result of the diagnostic information is correct, it will be stored in the Correct Consultation Experience Database (CorrectKB); if the diagnostic result of the diagnostic information is wrong, it will be abstractly summarized by the chain thinking reviewer and stored in the chain thinking experience database (ChainKB).

[0081] 3.3. When the next patient is consulted, the system retrieves similar cases from the experience database based on the patient’s background and condition to enhance the current recommendations for the next patient consultation.

[0082] The above system may refer to a medical system that integrates multiple intelligent agent roles.

[0083] It should be noted that in order to verify the effectiveness of each module in the embodiment of this application, an ablation experiment was conducted, that is, the above-mentioned residual discussion mode, the use of the correct consultation experience database and the chain thinking experience database were cancelled. The experimental results show that the residual discussion mode can effectively reduce information redundancy and improve the efficiency of consultation discussion; while the use of the correct consultation experience database and the chain thinking experience database significantly improves the system's diagnostic accuracy on similar cases.

[0084] In addition, the embodiment of the present application uses two public data sets, MedQA / PubMedQA, for testing, that is, the data in the above two public data sets are provided to each module in the embodiment of the present application for testing, and when the answers are consistent, the answer is determined to be correct, and finally the accuracy of all answers in the test set is counted. In this way, through a large number of experimental verifications, the effectiveness of each module in the embodiment of the present application in complex medical scenarios is proved, which can significantly improve the diagnostic accuracy and work efficiency of medical consultation.

[0085] See also Figure 3 , Figure 3 A schematic diagram of a medical consultation method provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the following steps are included:

[0086] Step 301: Obtain a target information set from a shared conversation sample pool, wherein the target information set includes M sample information combinations, and the sample information combination includes N rounds of consultation information. The N rounds of consultation information included in each sample information combination are consultation information between the same intelligent agent and the target patient, and different sample information combinations correspond to different intelligent agents. M and N are both integers greater than 1.

[0087] Step 302: Output the diagnostic information of the target patient according to the target information set.

[0088] As an optional implementation, each round of the N rounds of medical consultation information includes predicted diagnosis information, and outputting the diagnosis information of the target patient according to the target information set includes:

[0089] Acquire a target information from each sample information combination to obtain M target information, wherein the target information is the maximum predicted diagnosis information at the consultation time among the N rounds of consultation information included in each sample information combination;

[0090] In the case that all the M target information are matched, any one of the M target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output.

[0091] As an optional implementation, the outputting the diagnostic information of the target patient according to the target information set further includes:

[0092] In the case that there are L target information mismatches among the M target information, obtaining the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination, where L is an integer greater than or equal to 1, and L is less than M;

[0093] When the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination is greater than the preset number of rounds, determine the size of L and ML;

[0094] When L is greater than ML, any one of the L target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output;

[0095] In the case where L is less than ML, determining any one of the target information among the M target information except the L target information as the diagnosis information of the target patient, and outputting the diagnosis information of the target patient;

[0096] When L is equal to ML, any one of the M target information except the L target information, or any one of the L target information, or a randomly selected one of the M target information is determined as the diagnostic information of the target patient, and the diagnostic information of the target patient is output.

[0097] As an optional implementation manner, the inquiry information included in the N rounds of medical inquiry information is:

[0098] The query information in the sample case information obtained by the corresponding intelligent agent from the preset correct consultation experience database; and / or,

[0099] Inquiry information from sample case information obtained from the preset chain thinking experience database.

[0100] As an optional implementation, after outputting the diagnosis information of the target patient according to the target information set, the method further includes:

[0101] Sending the diagnosis information of the target patient to the first agent;

[0102] Wherein, the first agent is used to review the diagnostic information of the target patient, and if it is determined that the diagnostic information of the target patient is correct, store the diagnostic information of the target patient in the correct consultation experience database, and if it is determined that the diagnostic information of the target patient is wrong, send the diagnostic information of the target patient to the second agent;

[0103] The second intelligent agent is used to abstract and summarize the diagnostic information of the target patient, and store the target information obtained after the abstract summary in the chain thinking experience database. The target information includes error information existing in the diagnostic information of the target patient and suggestion information for modifying the error information.

[0104] It should be noted that the medical consultation method provided in the embodiment of the present application can be applied to the self-evolving medical consultation system based on multi-agent and large language model in the above-mentioned embodiment, and the various technical features in the medical consultation method can refer to the corresponding descriptions in the above-mentioned embodiment, and have the same beneficial technical effects, which will not be repeated here.

[0105] The present application also provides an electronic device. Figure 4 , the electronic device may include a processor 401, a memory 402, and a program 4021 stored in the memory 402 and executable on the processor 401. When the program 4021 is executed by the processor 401, Figure 3 Any steps in the corresponding method embodiments and achieving the same beneficial effects will not be repeated here.

[0106] A person skilled in the art can understand that all or part of the steps of implementing the above-mentioned embodiment method can be completed by hardware related to program instructions, and the program can be stored in a readable medium. The present application also provides a readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned Figure 3 Any steps in the corresponding method embodiments can achieve the same technical effect and will not be repeated here to avoid repetition.

[0107] The storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0108] The present application also provides a computer program product, including computer instructions, which can achieve the above-mentioned Figure 3 Any steps in the corresponding method embodiments can achieve the same technical effect and will not be repeated here to avoid repetition.

[0109] The above is a preferred implementation of the embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A self-evolving medical consultation system based on multi-agent and large language model, characterized in that: include: An acquisition module is used to acquire a target information set from a shared conversation sample pool, wherein the target information set includes M sample information combinations, wherein the sample information combination includes N rounds of consultation information, wherein the N rounds of consultation information included in each sample information combination are consultation information between the same agent and the target patient, and different sample information combinations correspond to different agents, and both M and N are integers greater than 1; An output module is used to output the diagnostic information of the target patient according to the target information set.

2. The self-evolving medical consultation system based on multi-agent and large language model according to claim 1 is characterized in that: Each round of medical consultation information in the N rounds of medical consultation information includes predicted diagnosis information, and the output module is further used to: Acquire a target information from each sample information combination to obtain M target information, wherein the target information is the maximum predicted diagnosis information at the consultation time among the N rounds of consultation information included in each sample information combination; In the case that all the M target information are matched, any one of the M target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output.

3. The self-evolving medical consultation system based on multi-agent and large language model according to claim 2 is characterized in that: The output module is further used for: In the case that there are L target information mismatches among the M target information, obtaining the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination, where L is an integer greater than or equal to 1, and L is less than M; When the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination is greater than the preset number of rounds, determine the size of L and ML; When L is greater than ML, any one of the L target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output; In the case where L is less than ML, determining any one of the target information among the M target information except the L target information as the diagnosis information of the target patient, and outputting the diagnosis information of the target patient; When L is equal to ML, any one of the M target information except the L target information, or any one of the L target information, or a randomly selected one of the M target information is determined as the diagnostic information of the target patient, and the diagnostic information of the target patient is output.

4. The self-evolving medical consultation system based on multi-agent and large language model according to any one of claims 1 to 3, characterized in that: The inquiry information included in the N rounds of medical consultation information is: The query information in the sample case information obtained by the corresponding intelligent agent from the preset correct consultation experience database; and / or, Inquiry information from sample case information obtained from the preset chain thinking experience database.

5. The self-evolving medical consultation system based on multi-agent and large language model according to any one of claims 1 to 3, characterized in that: Also includes: A sending module, used for sending the diagnosis information of the target patient to the first agent; Wherein, the first agent is used to review the diagnostic information of the target patient, and if it is determined that the diagnostic information of the target patient is correct, store the diagnostic information of the target patient in the correct consultation experience database, and if it is determined that the diagnostic information of the target patient is wrong, send the diagnostic information of the target patient to the second agent; The second intelligent agent is used to abstract and summarize the diagnostic information of the target patient, and store the target information obtained after the abstract summary in the chain thinking experience database. The target information includes error information existing in the diagnostic information of the target patient and suggestion information for modifying the error information.

6. A medical consultation method, characterized in that: include: Obtain a target information set from the shared conversation sample pool, wherein the target information set includes M sample information combinations, and the sample information combination includes N rounds of consultation information. The N rounds of consultation information included in each sample information combination are consultation information between the same agent and the target patient, and different sample information combinations correspond to different agents. M and N are both integers greater than 1. The diagnostic information of the target patient is output according to the target information set.

7. The method according to claim 6, characterized in that Each round of medical consultation information in the N rounds of medical consultation information includes predicted diagnosis information, and outputting the diagnosis information of the target patient according to the target information set includes: Acquire a target information from each sample information combination to obtain M target information, wherein the target information is the maximum predicted diagnosis information at the consultation time among the N rounds of consultation information included in each sample information combination; In the case that all the M target information are matched, any one of the M target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output.

8. The method according to claim 7, characterized in that Outputting the diagnosis information of the target patient according to the target information set also includes: In the case that there are L target information mismatches among the M target information, obtaining the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination, where L is an integer greater than or equal to 1, and L is less than M; When the number of consultation rounds corresponding to the N rounds of consultation information included in each sample information combination is greater than the preset number of rounds, determine the size of L and ML; When L is greater than ML, any one of the L target information is determined as the diagnosis information of the target patient, and the diagnosis information of the target patient is output; In the case where L is less than ML, determining any one of the target information among the M target information except the L target information as the diagnosis information of the target patient, and outputting the diagnosis information of the target patient; When L is equal to ML, any one of the M target information except the L target information, or any one of the L target information, or a randomly selected one of the M target information is determined as the diagnostic information of the target patient, and the diagnostic information of the target patient is output.

9. The method according to any one of claims 6 to 8, characterized in that The inquiry information included in the N rounds of medical consultation information is: The query information in the sample case information obtained by the corresponding intelligent agent from the preset correct consultation experience database; and / or, Inquiry information from sample case information obtained from the preset chain thinking experience database.

10. The method according to any one of claims 6 to 8, characterized in that After outputting the diagnosis information of the target patient according to the target information set, the method further includes: Sending the diagnosis information of the target patient to the first agent; Wherein, the first agent is used to review the diagnostic information of the target patient, and if it is determined that the diagnostic information of the target patient is correct, store the diagnostic information of the target patient in the correct consultation experience database, and if it is determined that the diagnostic information of the target patient is wrong, send the diagnostic information of the target patient to the second agent; The second intelligent agent is used to abstract and summarize the diagnostic information of the target patient, and store the target information obtained after the abstract summary in the chain thinking experience database. The target information includes error information existing in the diagnostic information of the target patient and suggestion information for modifying the error information.

Images