Disease diagnosis method, device, equipment and storage medium
By employing a multi-source knowledge-driven intelligent diagnostic method, combining a large language model and a disease-specific knowledge graph, and simulating multiple rounds of debate among multiple virtual doctors to verify medical guidelines, the problem of misdiagnosis and missed diagnosis in liver disease diagnosis is solved, achieving efficient and accurate liver disease diagnostic results.
Patent Information
- Application Number
- CN202411446867.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-16
- Publication Date
- 2025-10-17
- Estimated Expiration
- 2044-10-16
AI Technical Summary
The existing medical system suffers from problems such as high rates of misdiagnosis and missed diagnosis in liver disease diagnosis, complex and variable diagnosis, high requirements for doctors' knowledge, and uneven treatment levels among different medical institutions, resulting in low diagnostic accuracy and efficiency.
This approach employs a multi-source knowledge-driven intelligent diagnostic method. By acquiring patients' electronic medical record data, it analyzes abnormal medical indicators using large language models and disease-specific knowledge graphs. It simulates multiple rounds of debate among virtual doctors with different specialties and verifies the final results in conjunction with medical guidelines, ensuring the accuracy and interpretability of the diagnostic results.
It has improved the accuracy and efficiency of liver disease diagnosis, ensured consistency between diagnostic results and expert opinions, and enhanced the interpretability and scientific rigor of medical decisions.
Smart Images
Figure CN119339921B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of intelligent diagnosis, in particular to a multi-source knowledge driven intelligent disease diagnosis method, device, equipment and storage medium. BACKGROUND
[0002] The precise diagnosis and treatment of liver diseases greatly challenges the capacity of the existing medical system. Due to the diversity of liver diseases and the concealment of early symptoms, the lack of structured knowledge and systematic quantitative standards easily leads to misdiagnosis and missed diagnosis. Moreover, liver diseases have a long course and are often accompanied by multiple systemic lesions, requiring comprehensive discrimination of multi-disciplinary knowledge during the diagnosis process. These factors make liver disease diagnosis complex and variable, especially requiring a high level of knowledge breadth and depth from doctors. In addition, the diagnosis and treatment levels of different medical institutions are uneven, which further affects the decision-making accuracy and efficiency of complex liver diseases.
[0003] Therefore, it is urgent to develop and apply intelligent diagnosis methods to improve the accuracy and efficiency of complex liver disease diagnosis and treatment. Considering the seriousness of medical diagnosis, the most important thing for such a method is to have transparency and explainability, so that medical staff can understand and trust the diagnosis process and results, and thus be applied in clinical practice.
[0004] Therefore, the method of applying an intelligent liver disease diagnosis model to clinical liver disease analysis must meet the following conditions: 1. The method has rich knowledge of liver diseases and can deeply analyze the complex pathogenesis and differential methods of liver diseases to ensure the authenticity and scientificity of the output content; 2. The method needs to analyze all possible liver diseases to the greatest extent and give diagnostic basis to ensure that the results have high explainability and practicality; 3. The liver disease diagnosis result is highly similar to the result of expert doctors to ensure the accuracy of the model output and consistency with the level of experts. SUMMARY
[0005] In view of the above problems, the present application provides a disease diagnosis method, device, equipment and storage medium for overcoming the above problems or at least partially solving the above problems. The problem of deviation and error guidance of the diagnosis result caused by insufficient analysis of unstructured data and lack of professional knowledge in practical application is solved, which is a new method to realize the precision of disease diagnosis.
[0006] The present application provides the following solutions:
[0007] A disease diagnosis method, comprising:
[0008] obtaining an electronic medical record data set of a patient, dividing and counting the electronic medical record data set to obtain an original summary report;
[0009] abnormal medical indicators are obtained by analyzing the original summary report; the abnormal medical indicators and the names of the plurality of target diseases are taken as entities to search for related triples in a disease knowledge graph; a predefined analysis instruction is used to guide a large language model to explore the potential relationship between the abnormal medical indicators and the plurality of target diseases to generate corresponding diagnostic basis; the original summary report is fused with the related triples and the diagnostic basis to obtain a target data set after knowledge integration;
[0010] A plurality of agent roles of different tasks and levels are determined; a plurality of agent roles of the lowest level are used to generate a preliminary inference result in combination with the target data set; a plurality of agent roles of an intermediate level are used to generate an intermediate inference result in combination with the target data set and the preliminary inference result; a plurality of agent roles of the highest level are used to generate a final inference result in combination with the target data set, the preliminary inference result, and the generated intermediate inference result; a plurality of agent roles are used to vote on the final inference result for multiple rounds, and the final inference result is taken as a final argument result after the voting opinions are unified;
[0011] A guideline database including a plurality of target diseases and a consistency checking agent are determined; the consistency checking agent is used in combination with the guideline database and the original summary report to analyze the final argument result, so as to verify the accuracy of the final argument result according to the requirements of medical guidelines;
[0012] After the final argument result is determined to be accurate, a comprehensive diagnosis and treatment decision is generated.
[0013] Preferably, the original summary report at least includes patient history information, complaint information, image report, physical examination information, and laboratory examination result information.
[0014] Preferably, if the voting opinions are not unified, a modification suggestion is determined, so that a plurality of agent roles of the highest level adjust the final inference result according to the modification suggestion until the voting opinions are unified.
[0015] Preferably, the plurality of agent roles include 3 senior attending physician agents, 2 associate chief physician agents, and 1 chief physician agent from low to high levels;
[0016] The 3 senior attending physician agents are used in combination with the target data set to generate a preliminary inference result; the 2 associate chief physician agents are used in combination with the target data set and the preliminary inference result to generate an intermediate inference result; and the 1 chief physician agent is used in combination with the target data set, the preliminary inference result, and the generated intermediate inference result to generate a final inference result.
[0017] Preferably: the three senior attending physician intelligent agents generate three diagnostic conclusions from medical history, imaging examination and biochemical examination and fuse them to obtain the preliminary inference result.
[0018] Preferably: determining to use a prompt word strategy to divide the two deputy chief physician agents into an imaging and medical history comprehensive analysis agent and an individual assessment agent;
[0019] The imaging and medical history comprehensive analysis agent is used to review the patient's medical history, focusing on the timeline of disease progression and symptom changes, and compare current imaging results with past records to analyze the specific characteristics of the lesion;
[0020] The individual assessment agent is used to consider individual differences among patients and the impact of their lifestyle on the disease, evaluate changes in biomarkers in biochemical tests, and analyze the changing trends of these indicators at different time points.
[0021] Preferably, the target disease includes liver disease, the abnormal medical indicators include liver function indicators and related biomarkers and specific manifestations, and the liver function indicators at least include alanine aminotransferase content, aspartate aminotransferase content, and alkaline phosphatase content;
[0022] The guideline database includes at least primary liver cancer, drug-induced liver injury, fatty liver, liver cyst, liver hemangioma, autoimmune hepatitis, and alcoholic liver disease;
[0023] The comprehensive diagnosis and treatment decision at least includes the liver disease diagnosis conclusion, diagnostic basis and recommended treatment method.
[0024] A disease diagnosis device, for performing the above-mentioned disease diagnosis method, comprising:
[0025] A data acquisition unit is used to acquire a patient's electronic medical record data set, divide and count the electronic medical record data set to obtain an original summary report;
[0026] A knowledge enhancement unit is configured to obtain abnormal medical indicators by analyzing the original summary report; retrieve the abnormal medical indicators and the names of multiple target diseases as entities in a disease-specific knowledge graph to obtain related triples; use predefined analysis instructions to guide a large language model to explore the potential relationship between the abnormal medical indicators and the multiple target diseases to generate corresponding diagnostic evidence; and fuse the original summary report with the related triples and the diagnostic evidence to obtain a target data set after knowledge integration.
[0027] The multi-dimensional wisdom optimization unit is used to determine a plurality of agent roles of different tasks and levels; a preliminary inference result is generated by combining the target data set with the plurality of agent roles of the lowest level; an intermediate inference result is generated by combining the target data set and the preliminary inference result with the plurality of agent roles of the intermediate level; a final inference result is generated by combining the target data set, the preliminary inference result and the generated intermediate inference result with the plurality of agent roles of the highest level; and the final inference result is taken as a final argument result after a plurality of rounds of voting on the final inference result by the plurality of agent roles and the voting opinions are unified.
[0028] The self-correcting diagnosis and treatment collaborative verification unit is used to determine a guideline database including a plurality of target diseases and a consistency checking agent; the final argument result is analyzed by combining the guideline database and the original summary report with the consistency checking agent, so as to verify the accuracy of the final argument result according to the requirements of the medical guidelines;
[0029] The comprehensive diagnosis and treatment decision generation unit is used to generate a comprehensive diagnosis and treatment decision after the final argument result is accurate.
[0030] A disease diagnosis device, the device comprising a processor and a memory:
[0031] The memory is used to store program code and transmit the program code to the processor;
[0032] The processor is used to execute the disease diagnosis method according to the instructions in the program code.
[0033] A computer readable storage medium for storing program code, the program code being used to execute the disease diagnosis method.
[0034] According to the specific embodiments of the present application, the following technical effects are provided:
[0035] The disease diagnosis method, device, equipment and storage medium provided by the embodiments of the present application provide a disease intelligent diagnosis model to assist doctors in improving the accuracy and efficiency of clinical diagnosis. The problem of insufficient disease knowledge in general medical large models is solved, and the knowledge potential of the large model can be fully tapped. The model guides the accuracy of the pre-diagnosis result according to the current medical guidelines, and ensures the explainability of the medical decision. The model aims to improve the accuracy and efficiency of complex disease diagnosis by integrating multi-scale knowledge enhancement, multi-agent argument optimization and self-correcting diagnosis collaborative verification, and analyzes the scientificity of the conclusion based on the medical guidelines, and opens up a new path for auxiliary disease clinical medical diagnosis by combining the leading natural language processing technology and the clinical medical knowledge system.
[0036] Of course, implementing any product of the application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS
[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those of ordinary skill in the art without creative effort based on these drawings.
[0038] Figure 1 is a flowchart of a disease diagnosis method provided by an embodiment of the present application;
[0039] Figure 2 is a flowchart of a liver disease diagnosis process provided by an embodiment of the present application;
[0040] Figure 3 is a schematic diagram of a disease diagnosis device provided by an embodiment of the present application;
[0041] Figure 4 is a schematic diagram of a disease diagnosis device provided by an embodiment of the present application. DETAILED DESCRIPTION
[0042] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.
[0043] Referring to Figure 1 , a disease diagnosis method provided by an embodiment of the present application, as shown in Figure 1 , the method can include:
[0044] S101: Obtain an electronic medical record data set of a patient, divide and count the electronic medical record data set to obtain an original summary report; the electronic medical record data set can be manually input by a doctor, and a related database can also be established in advance, and medical records of patients who need to be diagnosed for a corresponding disease in a hospital are stored for subsequent diagnosis. The original summary can include some basic information, for example, in an implementation manner, the present application embodiment can provide that the original summary report at least includes patient history information, complaint information, image report, physical examination information and laboratory examination result information.
[0045] S102: obtaining an abnormal medical index by analyzing the original summary report; searching for relevant triples in a disease knowledge graph by taking the abnormal medical index and the names of a plurality of target diseases as entities; using a predefined analysis instruction to guide a large language model to explore the potential relationship between the abnormal medical index and a plurality of the target diseases to generate corresponding diagnostic basis; and fusing the original summary report, the relevant triples, and the diagnostic basis to obtain a target data set after knowledge integration;
[0046] S103: determining a plurality of agent roles of different tasks and levels; generating a preliminary inference result by using a plurality of agent roles of the lowest level in combination with the target data set; generating an intermediate inference result by using a plurality of agent roles of the intermediate level in combination with the target data set and the preliminary inference result; generating a final inference result by using a plurality of agent roles of the highest level in combination with the target data set, the preliminary inference result, and the generated intermediate inference result; determining a final argumentation result by using a plurality of agent roles to vote on the final inference result until the voting opinions are unified; and further, when the voting opinions are not unified, determining a modification suggestion, so that a plurality of agent roles of the highest level adjust the final inference result according to the modification suggestion until the voting opinions are unified.
[0047] The method provided by the embodiments of the present application uses a plurality of digital doctor agents with different expertise to simulate multi-round debates, and combines various opinions to form a preliminary diagnosis. The number and role positioning of the agents can be set according to the needs of the actual diagnosis scene. For example, in an implementation manner, the embodiments of the present application can further provide a plurality of agent roles including 3 senior attending physician agents, 2 deputy chief physician agents, and 1 chief physician agent from low to high levels.
[0048] The 3 senior attending physician agents generate a preliminary inference result in combination with the target data set; the 2 deputy chief physician agents generate an intermediate inference result in combination with the target data set and the preliminary inference result; and the 1 chief physician agent generates a final inference result in combination with the target data set, the preliminary inference result, and the generated intermediate inference result.
[0049] In a specific implementation, the embodiments of the present application can further provide that the 3 senior attending physician agents generate three diagnostic conclusions from the medical history, image examination, and biochemical examination, and fuse to obtain the preliminary inference result.
[0050] To further ensure that the method can focus on the multiple causes of the patient's illness, the embodiments of the present application can also provide a determination using a prompt word strategy to divide two said deputy chief physician agents into an imaging and medical history comprehensive analysis agent and an individual evaluation agent;
[0051] The imaging and medical history comprehensive analysis agent is used to review the patient's medical history, focus on the timeline of disease development and symptom changes, and compare current imaging results with past records to analyze the specific characteristics of the lesion;
[0052] The individual evaluation agent is used to consider the individual differences of the patient and the impact of their lifestyle on the disease, and to evaluate the changes in biomarkers in biochemical tests and analyze the trends of these indicators at different time points.
[0053] S104: Determine a guideline database including multiple said target diseases and a consistency checking agent; use the consistency checking agent to analyze the final argument result in combination with the guideline database and the original summary report, in order to verify the accuracy of the final argument result according to the requirements of the medical guidelines;
[0054] S105: Generate a comprehensive diagnosis and treatment decision after determining that the final argument result is accurate.
[0055] It can be understood that the method provided by the embodiments of the present application is applicable to the diagnosis of multiple diseases, and after determining the type of the target disease, corresponding abnormal medical indicators, data guideline databases, and comprehensive diagnosis and treatment decision output types, etc. can be set. For example, in one implementation, the embodiments of the present application can provide that the target disease includes liver disease, the abnormal medical indicators include liver function indicators and related biomarkers and specific manifestations, and the liver function indicators at least include alanine aminotransferase content, aspartate aminotransferase content, alkaline phosphatase content;
[0056] The guideline database at least includes primary liver cancer, drug-induced liver injury, fatty liver, liver cyst, liver hemangioma, autoimmune hepatitis, and alcoholic liver disease;
[0057] The comprehensive diagnosis and treatment decision at least includes a liver disease diagnosis conclusion, a diagnosis basis, and a recommended treatment method.
[0058] The disease diagnosis method provided by the embodiments of the present application improves the accuracy and reliability of diagnosis by integrating knowledge graph retrieval and large model generation technology, adopting a multi-agent debate strategy, and combining medical guidelines for liver diseases. Specifically, the method first analyzes abnormal medical indicators using multi-scale knowledge enhancement technology; secondly, it simulates multiple virtual doctors with different specializations to conduct multiple rounds of debate, synthesizes various viewpoints to form a preliminary diagnosis; finally, through a self-correcting diagnosis collaboration verification program, it ensures that the diagnosis result is consistent with the medical guidelines, thereby generating the final comprehensive diagnosis and treatment decision.
[0059] The disease diagnosis method provided by the embodiments of the present application is described in detail below, taking liver disease as an example. Referring to Figure 2 .
[0060] The method provided by the embodiments of the present application improves the accuracy and reliability of diagnosis by integrating knowledge graph retrieval and large model generation technology, adopting a multi-agent debate strategy, and combining medical guidelines for liver diseases. Specifically, the method first analyzes abnormal medical indicators using multi-scale knowledge enhancement technology, and constructs relevant triples by retrieving liver disease knowledge graph.
[0061] Secondly, it simulates multiple virtual doctors with different specializations to conduct multiple rounds of debate, synthesizes various viewpoints to form a preliminary diagnosis;
[0062] Finally, through a self-correcting diagnosis collaboration verification program, it ensures that the diagnosis result is consistent with the medical guidelines, thereby generating the final comprehensive diagnosis and treatment decision.
[0063] The multi-source knowledge driven intelligent liver disease diagnosis method includes the following steps:
[0064] First step: First read in the electronic medical record data set containing patient medical history information, imaging examination, biochemical examination, physical examination information, etc. Divide and count these information for subsequent processing.
[0065] Given a patient's electronic medical record, clean and integrate the patient's medical history, chief complaint, imaging report, physical examination and laboratory examination, etc. Free text information to form a summary report S.
[0066] Second step: Multi-scale knowledge enhancement. By analyzing the abnormal medical indicators in the electronic health record of the electronic medical record data set, and combining the retrieval of relevant triples knowledge in the liver disease knowledge graph, the understanding of the patient's liver disease condition is enhanced. This stage also includes intelligent analysis and generation, which guides the large language model to explore the potential association between abnormal medical indicators and multiple liver diseases through pre-defined instructions, and generates diagnosis basis. Finally, these retrieved and generated knowledge are fused with the original input data to realize multi-level knowledge integration.
[0067] By analyzing abnormal medical indicators in electronic medical records and retrieving relevant knowledge in the liver disease knowledge graph, a comprehensive analysis report is generated. Specific prompts are used to guide the LLM (Large Language Model) to analyze abnormal medical indicators (AMI) in medical text. These indicators reflect the individual's physiological state at a specific time point and include liver function indicators (such as alanine aminotransferase, aspartate aminotransferase, bilirubin, etc.) and other related biomarkers and specific manifestations. This process can be expressed as:
[0068] Gen AMI =LLM(S,prompt AMI )
[0069] Where: Gen AMI Indicates abnormal medical indicators, prompt AMI It means that it contains the role positioning and task prompts given to the large language model. LLM stands for large language model.
[0070] In order to ensure the professionalism and depth of knowledge enhancement, abnormal medical indicators and various liver disease names are used as entities to retrieve related triples Ret in the disease-specific knowledge graph. triples .
[0071] This disease-specific knowledge graph integrates liver disease clinical data, CM3KG, UMLS, and Wikidata5M, and contains approximately 140,000 nodes and relationships. The output format of the retrieval process can be expressed as:
[0072] Ret triples ={(HE i ,R i ,TE i )|i=1,…,n}
[0073] In the formula: HE i Represents the head entity, TE i Represents the tail entity, R i represents the relationship between them, and n is the number of triples retrieved.
[0074] In clinical practice, we emphasize the breadth and diversity of medical information to fully understand a patient's liver disease status and make a scientifically sound diagnosis. Using predefined analysis instructions, we can also guide large language models to explore the potential relationships between abnormal medical indicators (AMI) and various liver diseases and generate corresponding diagnostic evidence. This process can be formalized as follows:
[0075] Gen analysis =LLM(AMI,prompt analysis )
[0076] Where: Gen analysisrepresents the diagnosis basis, prompt analysis AMI represents abnormal medical indicators.
[0077] Then, we integrate the knowledge retrieved and generated with the original input data, achieving multi-level knowledge integration, which can be formalized as:
[0078] K = Merge (S, Ret triples , Gen analysis )
[0079] Where: K represents the target data set, and Merge represents the merge sort algorithm.
[0080] Step 3: Multi-dimensional wisdom optimization. Construct multiple intelligent agent roles (such as senior attending physicians, deputy chief physicians, and chief physicians) to analyze patient data from different perspectives, reveal and correct different cognitive biases, and improve the overall understanding of complex liver disease conditions. In each round of discussion, the intelligent agents vote on the current decision results, and if there is a disagreement, detailed modification suggestions must be proposed, followed by adjustment by the chief physician agent until the opinions are unified and the final debate result is generated.
[0081] Construct intelligent agent roles for different tasks to simulate the expert consultation process, conduct multiple rounds of debate and optimization on the case, and achieve consistency in the diagnosis conclusion.
[0082] Considering the complexity of liver disease and the characteristics of clinical diagnosis process, progressive medical roles are set for intelligent agents. The roles are divided into three levels: three senior attending physicians, two deputy chief physicians, and one chief physician. Each level of physician has different concerns when facing problems.
[0083] Specifically, j can be used to represent the different expertise areas of intelligent agents, including medical history analysis, imaging examination analysis, and biochemical examination analysis. The intelligent agent definition can be formalized as:
[0084] Agent j = LLM (prompt j ) j ∈ (MA, IA, BA, …)
[0085] Where: prompt j represents the intelligent agent prompt word engineering with different focuses, MA represents medical history analysis, IA represents imaging analysis, and BA represents biochemical examination.
[0086] Three senior attending physician agents generate three diagnosis conclusions from medical history, imaging examination, and biochemical examination, and fuse them to obtain the preliminary inference result Ans Pre . This process can be represented as:
[0087] AnsPre = Fusion j∈(MA,IA,BA) (f j (Agent j ,K))
[0088] where ∑ represents fusion, f j represents each agent generating diagnostic conclusions within its area of expertise.
[0089] Similarly, the Agent j = LLM(prompt j ) sets different focus directions for the associate chief physician agent through the prompt word strategy.
[0090] First, the imaging and medical history comprehensive analysis agent (Agent IM ) mainly reviews the patient's medical history, focuses on the timeline of disease progression and symptom changes, and compares the current imaging results with past records to analyze the specific characteristics of the lesion.
[0091] Second, the biochemical examination and individual assessment agent (Agent BI ) focuses on the individual differences of the patient and the impact of their lifestyle on the disease, while also evaluating the changes in biomarkers in biochemical examinations and analyzing the trends of these indicators at different time points.
[0092] Two comprehensive reports are generated by two associate chief physician agents, and they are integrated into an intermediate inference result Ans Mid . This process can be represented as:
[0093] Ans Mid = Fusion j∈(IM,BI) (f j (Agent j ,K,Ans Pre ))
[0094] The chief physician agent (Agent ID ) focuses on the integration and decision optimization direction. Its function is to integrate the analysis results of imaging, medical history, biochemical examination and individualized factors in Ans Mid . At the same time, considering the weight and reliability of all aspects of information to make a balanced judgment to avoid one-sided conclusions. We let it make high-level reasoning and decision-making based on comprehensive information and provide the final inference result Ans. This process can be represented as:
[0095] Ans = f ID (Agent ID ,K,Ans Mid )
[0096] where: f IDA function that represents the chief physician agent generating a conclusion.
[0097] In each round of discussion, all agents will vote on the current final inference result for up to 5 rounds (agree / disagree). If you disagree with the report, you need to make detailed modification suggestions Gen revise , which can be expressed as:
[0098]
[0099] In the formula: Fusion represents an adaptive fusion algorithm for merging multiple source modification suggestions into a final modification plan.
[0100] Then, the chief physician agent adjusts the report according to these modification suggestions Gen adjust , which can be expressed as:
[0101] Gen adjust = f ID (Agent ID , K, Ans, Gen revise )
[0102] Until the opinions are unified to generate the final debate result D n , which can be expressed as:
[0103]
[0104] In the formula: represents the modification suggestion prompt for different agents, Iterate(D n-1 , Gen adjust ) represents the iteration process of adjusting the last round of debate results D n-1 .
[0105] Step 4: Self-correcting diagnosis and treatment collaborative verification. This stage uses a guideline database containing multiple liver diseases as a basis for the reliability of intelligent decision-making. If the debate result is inconsistent with the medical guidelines, the intelligent agent will re-analyze the patient's condition in combination with the medical record and guidelines.
[0106] A guideline database Db Guide containing primary liver cancer, drug-induced liver damage, fatty liver, and other liver diseases is constructed as an important basis for verifying the reliability of intelligent decision-making.
[0107] An agreement verification intelligent agent Agent j is constructed through Agent j = LLM(prompt CV )The large language model is guided to analyze the multi-round debate results combined with medical guidelines and S, and to verify the accuracy of the debate results according to the requirements of the medical guidelines. If the conclusions are inconsistent, the agent will re-analyze the condition combined with the medical record and the guidelines.
[0108] Step 5: Generate a comprehensive diagnosis and treatment decision containing the diagnosis conclusion of liver disease, diagnosis basis and recommended treatment method. Generate a comprehensive diagnosis and treatment decision C containing the diagnosis conclusion of liver disease, diagnosis basis and recommended treatment method, as the output conclusion of the method, this process can be represented as:
[0109] C = f CV (Agent CV ,K,D n ,Db Guide )
[0110] In the formula: f CV represents a function of the consistency verification agent generating a conclusion, Db Guide represents a guideline database.
[0111] In summary, the disease diagnosis method provided in the present application provides a disease intelligent diagnosis model to assist doctors in improving the accuracy and efficiency of clinical diagnosis. It solves the problem of insufficient disease knowledge in general medical large models and can fully tap the knowledge potential of large models. The guided model verifies the accuracy of the pre-diagnosis result according to the current medical guidelines to ensure the explainability of the medical decision. The model aims to improve the accuracy and efficiency of complex disease diagnosis through the integration of multi-scale knowledge enhancement, multi-agent debate optimization and self-correcting diagnosis collaboration verification, and analyzes the scientificity of the conclusion based on medical guidelines, and opens up a new path for auxiliary disease clinical medical diagnosis by combining the cutting-edge natural language processing technology with the clinical medical knowledge system.
[0112] Referring to Figure 3 , the embodiments of the present application can also provide a disease diagnosis device, as shown in Figure 3 , for executing the disease diagnosis method described above, which can include:
[0113] A data acquisition unit 301 is configured to acquire an electronic medical record data set of a patient, divide and count the electronic medical record data set to obtain an original summary report;
[0114] A knowledge enhancement unit 302 is configured to obtain abnormal medical indicators by analyzing the original summary report; retrieve relevant triples in a special disease knowledge graph by taking the abnormal medical indicators and the names of a plurality of target diseases as entities; guide a large language model to explore the potential relationship between the abnormal medical indicators and a plurality of target diseases to generate corresponding diagnosis basis by using a predefined analysis instruction; and fuse the original summary report, the relevant triples and the diagnosis basis to obtain a knowledge-integrated target data set.
[0115] The multi-dimensional wisdom optimization unit 303 is configured to determine a plurality of agent roles of different tasks and levels, generate a preliminary inference result by using the agent roles of the lowest level in combination with the target data set, generate an intermediate inference result by using the agent roles of the intermediate level in combination with the target data set and the preliminary inference result, generate a final inference result by using the agent roles of the highest level in combination with the target data set, the preliminary inference result and the generated intermediate inference result, and determine the final inference result as a final argument result after a plurality of rounds of voting on the final inference result by using the agent roles.
[0116] The self-correction diagnosis and treatment collaborative verification unit 304 is configured to determine a guideline database including a plurality of target diseases and a consistency checking agent, and analyze the final argument result by using the consistency checking agent in combination with the guideline database and the original summary report, so as to verify the accuracy of the final argument result according to the requirements of the medical guidelines.
[0117] The comprehensive diagnosis and treatment decision generation unit 305 is configured to generate a comprehensive diagnosis and treatment decision after the final argument result is determined to be accurate.
[0118] The embodiments of the present application can also provide a disease diagnosis device, which comprises a processor and a memory.
[0119] The memory is configured to store program code and transmit the program code to the processor.
[0120] The processor is configured to execute the steps of the disease diagnosis method according to the instructions in the program code.
[0121] As shown in Figure 4 The disease diagnosis device provided by the embodiments of the present application can comprise a processor 10, a memory 11, a communication interface 12 and a communication bus 13. The processor 10, the memory 11 and the communication interface 12 can communicate with each other through the communication bus 13.
[0122] In the embodiments of the present application, the processor 10 can be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic devices, etc.
[0123] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiments of the disease diagnosis method.
[0124] The memory 11 stores one or more programs, which can include program codes including computer operation instructions. In the embodiments of the present application, the memory 11 at least stores programs for implementing the following functions:
[0125] Obtain an electronic medical record data set of a patient, divide and count the electronic medical record data set to obtain an original summary report;
[0126] Obtain an abnormal medical index by analyzing the original summary report; search for related triples in a special disease knowledge graph by taking the abnormal medical index and the names of a plurality of target diseases as entities; use a predefined analysis instruction to guide a large language model to explore the potential relationship between the abnormal medical index and a plurality of target diseases to generate corresponding diagnosis basis; and fuse the original summary report, the related triples, and the diagnosis basis to obtain a target data set after knowledge integration;
[0127] Determine a plurality of agent roles of different tasks and levels; use a plurality of agent roles of the lowest level to generate a preliminary inference result in combination with the target data set; use a plurality of agent roles of an intermediate level to generate an intermediate inference result in combination with the target data set and the preliminary inference result; use a plurality of agent roles of the highest level to generate a final inference result in combination with the target data set, the preliminary inference result, and the generated intermediate inference result; use a plurality of agent roles to vote on the final inference result for multiple rounds, and determine a final argument result after the voting opinions are unified;
[0128] Determine a guideline database including a plurality of target diseases and a consistency checking agent; use the consistency checking agent in combination with the guideline database and the original summary report to analyze the final argument result, so as to verify the accuracy of the final argument result according to the requirements of medical guidelines;
[0129] Determine the final argument result to be accurate to generate a comprehensive diagnosis and treatment decision.
[0130] In a possible implementation, the memory 11 can include a program storage area and a data storage area. The program storage area can store an operating system and application programs required by at least one function (such as a file creation function, a data read-write function), etc. The data storage area can store data created during use, such as initialization data, etc.
[0131] In addition, the memory 11 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device or other volatile solid-state storage device.
[0132] The communication interface 12 can be an interface of a communication module, used for connecting with other devices or systems.
[0133] Of course, it should be noted that, Figure 4 The structure shown does not constitute a limitation on the disease diagnosis device in the embodiments of the present application, and in actual applications, the disease diagnosis device can include more or fewer components than Figure 4 shown, or combine certain components.
[0134] The embodiments of the present application can also provide a computer readable storage medium for storing program codes, the program codes being used to execute the steps of the disease diagnosis method.
[0135] It should be noted that, in this document, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.
[0136] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in the various embodiments or some parts of the embodiments of the present application.
[0137] The various embodiments described in this specification are presented as examples of the application. Each example is provided by way of best mode, and variations of or additions to these examples can be possible. For example, the various embodiments described in this specification can be combined in different combinations. Further, other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. For example, to implement a system embodiment, one can implement a method embodiment and one or more system modules to perform the method embodiment. Each of the various embodiments can be implemented alone or in combination with any other embodiments. It is therefore intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
[0138] Any modification, equivalent replacement, improvement, and the like made within the spirit and principle of the application shall fall within the scope of the protection of the application.
Claims
1. A disease diagnosis method, characterized in that: include: Obtaining a patient's electronic medical record data set, dividing and statistically analyzing the electronic medical record data set to obtain an original summary report; Obtaining abnormal medical indicators by analyzing the original summary report; The abnormal medical indicators and the names of multiple target diseases are used as entities to retrieve relevant triples in the disease-specific knowledge graph; predefined analysis instructions are used to guide the large language model to explore the potential relationship between the abnormal medical indicators and the multiple target diseases to generate corresponding diagnostic evidence; the original summary report is integrated with the relevant triples and the diagnostic evidence to obtain a target data set after knowledge integration; Determine a plurality of intelligent agent roles with different tasks and progressive levels; generate a preliminary inference result using the lowest-level intelligent agent roles in combination with the target data set; generate an intermediate inference result using the intermediate-level intelligent agent roles in combination with the target data set and the preliminary inference result; and generate a final inference result using the highest-level intelligent agent roles in combination with the target data set, the preliminary inference result, and the generated intermediate inference result; Using a plurality of the intelligent agent roles to conduct multiple rounds of voting on the final inference result, and after determining that the voting opinions are unified, the final inference result is used as the final debate result; determining a guideline database including a plurality of target diseases and a consistency checking agent; Utilizing the consistency checking agent to analyze the final debate result in combination with the guideline database and the original summary report, so as to verify the accuracy of the final debate result according to the requirements of the medical guidelines; After confirming that the final debate result is accurate, a comprehensive diagnosis and treatment decision is generated; The plurality of agent roles include, from low to high levels, three senior attending physician agents, two deputy chief physician agents, and one chief physician agent; The three senior attending physician agents generated three diagnostic conclusions based on medical history, imaging examination, and biochemical examination and integrated them to obtain the preliminary inference result; Determine to use the prompt word strategy to divide the two deputy chief physician agents into an imaging and medical history comprehensive analysis agent and an individual assessment agent; The imaging and medical history comprehensive analysis agent is used to review the patient's medical history, focusing on the timeline of disease progression and symptom changes, and compare current imaging results with past records to analyze the specific characteristics of the lesion; The individual assessment agent is used to consider individual differences in patients and the impact of their lifestyle on the disease, and to evaluate changes in biomarkers in biochemical tests and analyze the changing trends of these indicators at different time points; The target disease includes liver disease, and the abnormal medical indicators include liver function indicators and related biomarkers and specific manifestations, and the liver function indicators at least include alanine aminotransferase level, aspartate aminotransferase level, and alkaline phosphatase level; The guideline database includes at least primary liver cancer, drug-induced liver injury, fatty liver, liver cyst, liver hemangioma, autoimmune hepatitis, and alcoholic liver disease; The comprehensive diagnosis and treatment decision at least includes the liver disease diagnosis conclusion, diagnostic basis and recommended treatment method.
2. The disease diagnosis method according to claim 1, characterized in that: The original summary report includes at least the patient's medical history information, chief complaint information, imaging report, physical examination information and laboratory test result information.
3. The disease diagnosis method according to claim 1, characterized in that: After determining that the voting opinions are not unified, a modification suggestion is determined so that the highest-level intelligent agent roles adjust the final inference result according to the modification suggestion until the voting opinions are unified.
4. A disease diagnosis device, characterized in that: For performing the disease diagnosis method according to any one of claims 1 to 3, the device comprises: A data acquisition unit is used to acquire a patient's electronic medical record data set, divide and count the electronic medical record data set to obtain an original summary report; A knowledge enhancement unit is configured to obtain abnormal medical indicators by analyzing the original summary report; retrieve the abnormal medical indicators and the names of multiple target diseases as entities in a disease-specific knowledge graph to obtain related triples; use predefined analysis instructions to guide a large language model to explore the potential relationship between the abnormal medical indicators and the multiple target diseases to generate corresponding diagnostic evidence; and fuse the original summary report with the related triples and the diagnostic evidence to obtain a target data set after knowledge integration. A multi-dimensional intelligent debate optimization unit is configured to determine a plurality of intelligent agent roles with different tasks and progressive levels; generate a preliminary inference result by combining the target data set with the intelligent agent roles at the lowest level; generate an intermediate inference result by combining the target data set and the preliminary inference result with the intelligent agent roles at the intermediate level; generate a final inference result by combining the target data set, the preliminary inference result, and the generated intermediate inference result with the intelligent agent roles at the highest level; and perform multiple rounds of voting on the final inference result by the intelligent agent roles, and use the final inference result as the final debate result after unanimous voting opinions are determined. a self-correcting diagnosis and treatment collaborative verification unit, configured to determine a guideline database including a plurality of target diseases and a consistency verification agent; and to analyze the final debate result using the consistency verification agent in combination with the guideline database and the original summary report, so as to verify the accuracy of the final debate result in accordance with the requirements of the medical guidelines; A comprehensive diagnosis and treatment decision generating unit is used to generate a comprehensive diagnosis and treatment decision after determining that the final debate result is accurate.
5. A disease diagnosis device, characterized in that: The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the disease diagnosis method according to any one of claims 1 to 3 according to the instructions in the program code.
6. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store program code, and the program code is used to execute the disease diagnosis method according to any one of claims 1 to 3.