Chronic disease-oriented digital medical AI large model diagnosis and treatment information retrieval method and system

Abstract

The invention provides a chronic disease-oriented digital medical AI large model diagnosis and treatment information retrieval method and system. The method comprises the following steps: constructing chronic disease medical text data; performing fine tuning training on the general language model by utilizing the chronic disease medical text data to obtain a medical large model; acquiring to-be-diagnosed and treated disease data of the patient; performing cue word optimization and retrieval by using an intelligent agent to obtain an RAG disease map data text; and converting the RAG disease graph data text into to-be-diagnosed and treated disease data in a knowledge graph form by adopting a knowledge graph embedding technology, inputting the to-be-diagnosed and treated disease data into a graph neural network to obtain the to-be-diagnosed and treated disease data for realizing knowledge reasoning, inputting the to-be-diagnosed and treated disease data into a medical big model, generating text output, and obtaining an inquiry service health analysis report. The technical effects of improving the retrieval efficiency and the generation process, enhancing the understanding of complex knowledge, overcoming the defect of insufficient disease data sources, providing a rapid, efficient and accurate diagnosis and treatment scheme for the patient, enhancing the self-health management ability of the patient and relieving the medical resource pressure are achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical information retrieval technology, and in particular to a digital medical AI large-scale model diagnosis and treatment information retrieval method and system for chronic diseases. Background Art

[0002] With the rapid development of artificial intelligence (AI), the popularity of large language models (LLMs), particularly the GPT series developed by OpenAI, has become increasingly important across various industries. In particular, these large language models have demonstrated tremendous potential in the healthcare sector, bringing numerous changes and breakthroughs to the industry. Numerous innovative AI models, specifically developed for healthcare scenarios, have emerged, solving practical medical problems and improving diagnostic accuracy, personalized treatment, and overall healthcare efficiency.

[0003] Currently, large-scale language models in the medical industry are primarily developed for specific fields and small scenarios. These models typically focus on a specific medical field or diagnosis and treatment scenario, with relatively limited functionality and coverage. Primarily focused on answering questions about diagnosis and treatment, these models primarily address questions arising within that specific scenario and are trained on a limited corpus. The technical architecture of these models generally utilizes a Transformer network structure, modeling and optimizing through deep learning. These models can provide valuable decision support in specific medical scenarios.

[0004] However, in the process of implementing the technical solutions in the embodiments of the present application, the inventors of the present application discovered that the above-mentioned prior art has at least the following technical problems:

[0005] Existing medical NLP models generally rely on the large language model interface provided by OpenAI. Their diagnosis and treatment recommendations are usually limited to specific diseases or symptoms. When faced with inquiries about chronic diseases, there are technical problems such as difficulty in conducting comprehensive and integrated analysis and difficulty in meeting complex clinical decision-making needs. Summary of the Invention

[0006] The embodiment of the present invention provides a digital medical AI large-model diagnosis and treatment information retrieval method and system for chronic diseases, which is used to solve the problem that medical NLP models in the existing technology generally rely on the large language model interface provided by OpenAI, and their diagnosis and treatment recommendations are usually limited to specific diseases or symptoms. When facing chronic disease consultations, there are technical problems such as difficulty in conducting comprehensive and integrated analysis and difficulty in meeting complex clinical decision-making needs. Based on the GLM pre-training framework, combined with the generation ability of the autoregressive model and the efficient information filling ability of the blank filling mechanism, the model can flexibly capture contextual information and fill in missing parts when processing complex texts, achieving the technical effect of improving retrieval efficiency and generation process, enhancing understanding of complex knowledge, overcoming insufficient disease data sources, providing patients with fast, efficient and accurate diagnosis and treatment plans, enhancing patients' self-health management capabilities, and alleviating pressure on medical resources.

[0007] In view of the above problems, an embodiment of the present application is proposed to provide a digital medical AI large model diagnosis and treatment information retrieval method and system for chronic diseases.

[0008] In the first aspect, the present invention provides a digital medical AI big model diagnosis and treatment information retrieval method for chronic diseases, the method comprising: obtaining historical chronic disease medical data and constructing chronic disease medical text data; obtaining a general language model GLM-4-9B-Chat; using the chronic disease medical text data to fine-tune the general language model GLM-4-9B-Chat to obtain a medical big model GLM-4LocalDisease; obtaining the patient's disease data to be diagnosed; using an intelligent agent to optimize and retrieve prompt words for the disease data to be diagnosed to obtain RAG disease graph data text; using knowledge graph embedding technology to convert the RAG disease graph data text into the disease data to be diagnosed in the form of a knowledge graph; inputting the disease data to be diagnosed in the form of a knowledge graph into a graph neural network to obtain the disease data to be diagnosed that implements knowledge reasoning; inputting the disease data to be diagnosed that implements knowledge reasoning into the medical big model GLM-4LocalDisease to generate text output and obtain a health analysis report for a medical consultation service.

[0009] Preferably, the historical chronic disease medical data includes nearly 340,000 question and answer data in the fields of internal medicine, surgery, and pediatrics.

[0010] Preferably, obtaining historical chronic disease medical data and constructing chronic disease medical text data include: obtaining historical chronic disease medical data; classifying and labeling the historical chronic disease medical data according to the type of chronic disease, associating patient symptoms, treatment plans and specific drugs together to obtain labeled historical chronic disease medical data, wherein the chronic disease types are divided into surgery, internal medicine and pediatrics; cleaning and preprocessing the labeled historical chronic disease medical data to obtain chronic disease medical text data.

[0011] Preferably, the labeled historical chronic disease medical data is cleaned and preprocessed to obtain chronic disease medical text data, including: eliminating duplicate records in the historical chronic disease medical data to obtain first historical chronic disease medical data; performing missing value filling processing on the first historical chronic disease medical data to obtain second historical chronic disease medical data; deleting irrelevant text information in the second chronic disease medical data to obtain third historical chronic disease medical data; performing format unification processing on the third historical chronic disease medical data to obtain fourth historical chronic disease medical data, wherein the fourth historical chronic disease medical data meets the input data requirements of the general language model GLM-4-9B-Chat; and constructing chronic disease medical text data based on the fourth historical chronic disease medical data.

[0012] Preferably, the use of an intelligent agent to optimize and retrieve prompt words for the disease data to be diagnosed and treated to obtain RAG disease graph data text includes: judging whether the disease data to be diagnosed and treated has duplication problems based on contextual knowledge memory query; if the disease data to be diagnosed and treated has duplication problems, directly outputting it to the patient; setting an interface function according to the disease data to be diagnosed and treated, wherein the interface function includes age, gender, key knowledge, and problem optimization interface function; according to the interface function, using the intelligent agent to quickly parse the disease data to be diagnosed and treated and extract key information; setting priority according to the key information; according to the priority, calling different tool interfaces to optimize and process the disease data to be diagnosed and treated to obtain RAG disease graph data text.

[0013] Preferably, the knowledge graph embedding technology is used to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph, including: using the Wikidata data storage structure to make the RAG disease graph data text into a knowledge graph, and the structural triple relationship of the knowledge graph can be expressed as (A, is_friend_of, B), where A and B are entities and "is_friend_of" is a relationship; converting the entities and relationships in the knowledge graph into feature vectors of nodes and edges, where each node is represented as a feature vector, and each edge requires a corresponding feature representation, and the feature representation is generated by the relationship embedding; constructing an adjacency matrix, which represents the connection relationship between nodes and reflects the existence relationship between entities; obtaining graph structure data based on the node features, the edge features and the connection relationship; obtaining disease data to be diagnosed and treated in the form of a knowledge graph based on the graph structure data.

[0014] Preferably, the step of inputting the data on the disease to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the data on the disease to be diagnosed and treated that implements knowledge reasoning includes: inputting the data on the disease to be diagnosed and treated in the form of the knowledge graph into the graph neural network; based on the graph convolution mechanism, the graph neural network uses the node features and the edge features to learn the deep-level relationships between the entities; and obtaining the data on the disease to be diagnosed and treated that implements knowledge reasoning based on the deep-level relationships between the entities.

[0015] In a second aspect, the present invention provides a digital medical AI large-model diagnosis and treatment information retrieval system for chronic diseases, the system comprising:

[0016] A first construction unit, the first construction unit is used to obtain historical chronic disease medical data and construct chronic disease medical text data;

[0017] A first obtaining unit, wherein the first obtaining unit is used to obtain a general language model GLM-4-9B-Chat;

[0018] A second obtaining unit, the second obtaining unit is used to fine-tune the general language model GLM-4-9B-Chat using the chronic disease medical text data to obtain a large medical model GLM-4Local Disease;

[0019] a third obtaining unit, configured to obtain data of the patient's disease to be diagnosed and treated;

[0020] a fourth obtaining unit, configured to optimize and retrieve prompt words for the disease data to be diagnosed and treated using an intelligent agent to obtain RAG disease graph data text;

[0021] A first conversion unit, configured to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph using a knowledge graph embedding technology;

[0022] a fifth obtaining unit, configured to input the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated for implementing knowledge reasoning;

[0023] The first forming unit is used to input the disease data to be diagnosed and treated for realizing knowledge reasoning into the medical large model GLM-4 Local Disease, generate text output, and form a consultation service health analysis report.

[0024] In the third aspect, the present invention also provides a digital medical AI large model diagnosis and treatment information retrieval device for chronic diseases, comprising a memory, a processor, and a computer program stored on the memory and runnable on the processor, characterized in that when the computer program is executed by the processor, the processor executes the steps of the aforementioned digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases.

[0025] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, it is used to implement the steps of the aforementioned digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases.

[0026] The above one or more technical solutions in the embodiments of the present application have at least one or more of the following technical effects:

[0027] An embodiment of the present invention provides a digital medical AI large model diagnosis and treatment information retrieval method and system for chronic diseases, the method comprising: obtaining historical chronic disease medical data and constructing chronic disease medical text data; obtaining a general language model GLM-4-9B-Chat; using the chronic disease medical text data to fine-tune the general language model GLM-4-9B-Chat to obtain a medical large model GLM-4LocalDisease; obtaining the patient's disease data to be diagnosed and treated; using an intelligent agent to optimize and retrieve prompt words for the disease data to be diagnosed and treated to obtain RAG disease graph data text; using knowledge graph embedding technology to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph; inputting the disease data to be diagnosed and treated in the form of a knowledge graph into a graph neural network to obtain disease data to be diagnosed and treated that implements knowledge reasoning; inputting the disease data to be diagnosed and treated that implements knowledge reasoning into the medical large model GLM-4 Local Disease to generate text output and obtain a health analysis report for a medical consultation service. This technology aims to address the problem that existing medical NLP models generally rely on the large language model interface provided by OpenAI. Their diagnosis and treatment recommendations are usually limited to specific diseases or symptoms. When faced with consultations on chronic diseases, there are technical issues such as difficulty in conducting comprehensive and integrated analysis and difficulty in meeting the needs of complex clinical decision-making. Based on the GLM pre-training framework, combined with the generation capabilities of the autoregressive model and the efficient information filling capabilities of the blank filling mechanism, the model can flexibly capture contextual information and fill in missing parts when processing complex texts, achieving the technical effect of improving retrieval efficiency and generation process, enhancing the understanding of complex knowledge, overcoming the lack of disease data sources, providing patients with fast, efficient and accurate diagnosis and treatment plans, enhancing patients' ability to manage their own health, and alleviating the pressure on medical resources.

[0028] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of a method for retrieving diagnosis and treatment information of a digital medical AI large model for chronic diseases according to an embodiment of the present invention;

[0030] Figure 2 This is a flow chart of a digital medical AI large-model diagnosis and treatment information retrieval device for chronic diseases in an embodiment of the present invention;

[0031] Figure 3 This is an execution flow chart of an intelligent agent in an embodiment of the present invention;

[0032] Figure 4 This is an entity mapping relationship space diagram in an embodiment of the present invention;

[0033] Figure 5 This is a network structure diagram of the diagnosis and treatment information retrieval system of the digital medical AI large model in an embodiment of the present invention;

[0034] Figure 6 This is a diagram showing the operation results of the medical large model health assistant online system in an embodiment of the present invention.

[0035] Description of reference numerals: bus 300 , receiver 301 , processor 302 , transmitter 303 , memory 304 , bus interface 306 . DETAILED DESCRIPTION

[0036] The embodiments of the present invention provide a digital medical AI large-model diagnosis and treatment information retrieval method and system for chronic diseases, which solves the technical problem that medical NLP models in the existing technology generally rely on the large language model interface provided by OpenAI. Their diagnosis and treatment recommendations are usually limited to specific diseases or symptoms. When facing chronic disease consultations, there are difficulties in conducting comprehensive comprehensive analysis and meeting complex clinical decision-making needs.

[0037] The overall idea of ​​the technical solution provided by the present invention is as follows: historical chronic disease medical data is obtained to construct chronic disease medical text data; a general language model GLM-4-9B-Chat is obtained; the general language model GLM-4-9B-Chat is fine-tuned and trained using the chronic disease medical text data to obtain a large medical model GLM-4Local Disease; the patient's disease data to be diagnosed and treated is obtained; an intelligent agent is used to optimize and retrieve prompt words for the disease data to be diagnosed and treated to obtain RAG disease graph data text; knowledge graph embedding technology is used to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph; the disease data to be diagnosed and treated in the form of a knowledge graph is input into a graph neural network to obtain disease data to be diagnosed and treated that implements knowledge reasoning; the disease data to be diagnosed and treated that implements knowledge reasoning is input into the large medical model GLM-4LocalDisease to generate text output to obtain a health analysis report for a medical consultation service. Based on the GLM pre-training framework, combined with the generation ability of the autoregressive model and the efficient information filling ability of the blank filling mechanism, the model can flexibly capture contextual information and fill in missing parts when processing complex texts, achieving the technical effect of improving retrieval efficiency and generation process, enhancing the understanding of complex knowledge, overcoming the lack of disease data sources, providing patients with fast, efficient and accurate diagnosis and treatment plans, enhancing patients' self-health management capabilities, and alleviating the pressure on medical resources.

[0038] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. Unless there is a conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0039] The term "and / or" in this document simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0040] Example 1

[0041] Figure 1 Schematic diagram of a digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases in an embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases, the method comprising:

[0042] Step s1: Obtain historical chronic disease medical data and construct chronic disease medical text data;

[0043] Step s2: Obtain the general language model GLM-4-9B-Chat;

[0044] Step s3: fine-tuning the general language model GLM-4-9B-Chat using the chronic disease medical text data to obtain the large medical model GLM-4Local Disease;

[0045] Furthermore, the historical chronic disease medical data includes nearly 340,000 question and answer data in the fields of internal medicine, surgery, and pediatrics.

[0046] Furthermore, the acquisition of historical chronic disease medical data and the construction of chronic disease medical text data include:

[0047] Obtain historical chronic disease medical data;

[0048] Classify and annotate the historical chronic disease medical data according to the chronic disease type, and associate the patient's symptoms, treatment plan, and specific medication to obtain annotated historical chronic disease medical data, wherein the chronic disease type is divided into surgery, internal medicine, and pediatrics;

[0049] The annotated historical chronic disease medical data is cleaned and preprocessed to obtain chronic disease medical text data.

[0050] Furthermore, the aforementioned cleaning and preprocessing of the annotated historical chronic disease medical data to obtain chronic disease medical text data includes:

[0051] Eliminating duplicate records in the historical chronic disease medical data to obtain first historical chronic disease medical data;

[0052] Performing missing value filling processing on the first historical chronic disease medical data to obtain second historical chronic disease medical data;

[0053] Deleting irrelevant text information in the second chronic disease medical data to obtain third historical chronic disease medical data;

[0054] Performing format unification processing on the third historical chronic disease medical data to obtain fourth historical chronic disease medical data, wherein the fourth historical chronic disease medical data meets the input data requirements of the general language model GLM-4-9B-Chat;

[0055] Chronic disease medical text data is constructed based on the fourth historical chronic disease medical data.

[0056] Specifically, the embodiment of the present application provides a pre-training framework based on autoregressive blank filling, combining a general language model (GLM) of an autoregressive model and a blank filling mechanism, wherein the general language model is intended to improve the ability to capture contextual information when generating text, and can effectively fill in missing words or characters, thereby improving the fluency and accuracy of the generated text. Based on the GLM, it has shown higher generalization ability and flexibility in multiple task categories, which can not only efficiently understand and generate natural language, but also show significant advantages in the diagnosis and question-answering services of chronic diseases in the medical field. Through this pre-training framework, the model can span different fields, tasks and application scenarios to provide more accurate and efficient solutions. In the large-scale model medical assistant for chronic diseases of the present invention, targeted suggestions can be generated based on the symptoms and health data of different patients, thereby effectively improving the quality and efficiency of medical services. Specifically, the embodiment of the present application obtains historical chronic disease medical data, that is, self-built nearly 340,000 question-answer data covering internal medicine, surgery and pediatrics as RAG (Retrieval Augmented Generation) medical text data, thereby constructing chronic disease medical text data. In this embodiment of the present invention, the general language model GLM-4-9B-Chat is fine-tuned using historical chronic disease medical data to generate the large medical model GLM-4Local Disease, which is deployed on a local inference server. The key to the fine-tuning process lies in data engineering. The specific process involves collecting historical chronic disease medical data, classifying it according to different chronic disease types (surgery, internal medicine, and pediatrics), annotating symptoms and corresponding treatment plans, and associating specific medications with patients' conditions and symptoms. Because the raw data of historical chronic disease medical data may contain noise or missing values, the annotated data must be cleaned and preprocessed. This involves removing duplicate records, filling missing values, removing irrelevant text information, and unifying data in different formats to ensure that the data input to the general language model GLM-4-9B-Chat meets the requirements so that the general language model GLM-4-9B-Chat can correctly understand and process it. This invention further ensures the system's comprehensive understanding and response capabilities to diseases by independently constructing a comprehensive chronic disease medical text data set containing 340,000 data items covering internal medicine, surgery, pediatrics and other fields, avoiding the limitations of insufficient or incomplete external data sources, thereby ensuring that the health assistant's diagnosis and treatment recommendations are more accurate and comprehensive.

[0057] Step s4: obtaining the patient's disease data to be diagnosed and treated;

[0058] Step s5: using an intelligent agent to optimize and retrieve prompt words for the disease data to be diagnosed and treated, and obtain RAG disease graph data text;

[0059] Specifically, prompt words are commands that drive the operation of the large language model, which mainly uses predefined templates to allow the agent to replace the variable part in the template. By designing reasonable and accurate input prompt words, the medical large model GLM-4LocalDisease can be guided to generate outputs that better meet the needs. Traditional prompt word design relies on manual experience, while the use of the agent can improve retrieval efficiency and effectiveness through automated exploration, optimization and adjustment of prompt words. For some complex problems, directly handing them over to the medical large model GLM-4LocalDisease for answers may generate inaccurate, outdated or inappropriate medical information for specific situations. Therefore, the embodiment of the present application optimizes and retrieves the prompt words of the disease data to be diagnosed by the agent to obtain the RAG disease graph data text. Specifically, the agent Agent includes three modules: context memory module, tool list module and execution module, each with a clear division of labor, such as Figure 3 As shown, the collaborative work of these three modules can ensure that when the system handles medical problems, it can not only provide rich, complete and correct data input for the model, but also maintain efficiency and consistency in multiple rounds of patient interactions.

[0060] In step s5, the intelligent agent optimizes and retrieves the prompt words of the disease data to be diagnosed and treated to obtain the RAG disease graph data text, which specifically includes:

[0061] Based on contextual knowledge memory query, determine whether the disease data to be diagnosed and treated has duplication issues;

[0062] If there is any duplication in the disease data to be diagnosed and treated, it is directly output to the patient;

[0063] According to the disease data to be diagnosed and treated, an interface function is set, wherein the interface function includes age, gender, key knowledge, and problem optimization interface functions. Among them, these interface functions of Age(), Gender(), KeyKnowledge(), and Problem Optimization() can help the intelligent agent quickly parse user questions and extract key information, thereby improving retrieval efficiency.

[0064] According to the interface function, the intelligent agent is used to quickly analyze the disease data to be diagnosed and treated and extract key information;

[0065] Setting priorities based on the key information;

[0066] According to the priority, different tool interfaces are called to optimize and process the data of the disease to be diagnosed and treated to obtain RAG disease graph data text.

[0067] Step s6: Using knowledge graph embedding technology, convert the RAG disease graph data text into the disease data to be diagnosed and treated in the form of knowledge graph;

[0068] Furthermore, the knowledge graph embedding technology is used to convert the RAG disease graph data text into the disease data to be diagnosed and treated in the form of a knowledge graph, including:

[0069] The RAG disease graph data text is made into a knowledge graph using the Wikidata data storage structure. The structural triple relationship of the knowledge graph can be expressed as (A, is_friend_of, B), where A and B are entities and "is_friend_of" is a relationship;

[0070] Converting entities and relationships in the knowledge graph into feature vectors of nodes and edges, where each node is represented as a feature vector and each edge requires a corresponding feature representation generated by embedding the relationship;

[0071] Constructing an adjacency matrix, wherein the adjacency matrix represents the connection relationship between nodes and reflects the existence relationship between entities;

[0072] Obtaining graph structure data according to the node features, the edge features, and the connection relationship;

[0073] According to the graph structure data, data on diseases to be diagnosed and treated are obtained in the form of a knowledge graph.

[0074] Specifically, knowledge graph embedding (KGE) maps entities and relationships in a knowledge graph into a low-dimensional vector space, typically in the form of triples (subject, predicate, object), so that they can capture semantic information in the vector space. The basic principle is to represent entities and relationships by designing embedding vectors and optimize the representation of triples through a loss function, so that correct triples have higher similarity. Common optimization methods include gradient descent and negative sampling. Embedding vectors can be used for tasks such as entity linking, relationship prediction, and question-answering systems. Through vectorized representation, they improve the reasoning and query capabilities of knowledge graphs.

[0075] The structural triple relationship of the knowledge graph can be expressed as (A, is_friend_of, B), where A and B are entities and "is_friend_of" is a relationship. The goal of knowledge graph embedding is to convert these triplets into low-dimensional vectors so that the semantic information between entities and relationships can be captured in the vector space. The design of the embedding space makes similar entities and relationships closer in the vector space. For each entity and relationship, it is mapped to a low-dimensional vector space as shown in (Formula 1), and then the DistMult loss function is used to measure the error of the triple. This loss function is based on a model of the scoring function and calculates the inner product of the entity vector and the relationship matrix. The calculation formula is shown in Formula 2:

[0076] e→v e &r→v r (Formula 1)

[0077] Among them, e is an entity, r is a relation, ve and vr are the vector representations of entity e and relation r respectively.

[0078]

[0079] Where h is the head entity (head), r is the relation, t is the tail entity (tail), vh and vt are the embedding vectors of the head entity and the tail entity respectively. Mr is the matrix associated with the relation r, which represents the effect or mapping of the relation on the entity in the embedding space. This matrix is ​​usually a trainable parameter matrix. The transpose of the vector represents the row vector in the matrix multiplication. The relationship matrix Mr is used to map the head entity vector vh to the tail entity vector vt. By calculating the product of the head entity vector and the relationship matrix, and then performing the inner product operation with the tail entity vector, the score of the triple is obtained. This score represents the degree of matching of the triple (h, r, t). By optimizing this matrix, the structure of the relationship can be effectively captured in the embedding space. The entity mapping relationship space diagram is as follows: Figure 4 shown.

[0080] Step s7: inputting the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated that implements knowledge reasoning;

[0081] Step s8: Input the disease data to be diagnosed and treated for knowledge reasoning into the medical large model GLM-4LocalDisease, generate text output, and obtain a consultation service health analysis report.

[0082] Furthermore, the step of inputting the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated for implementing knowledge reasoning includes:

[0083] Inputting the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network;

[0084] Based on the graph convolution mechanism, the graph neural network uses the node features and the edge features to learn the deep relationships between the entities;

[0085] According to the deep-level relationships between the entities, the data of the disease to be diagnosed and treated for realizing knowledge reasoning is obtained.

[0086] Specifically, the embodiment of the present application uses nearly 340,000 question and answer data, that is, the chronic disease medical text data is made into a knowledge graph using Wikidata, and the knowledge graph is converted into data that can be recognized by the graph neural network (GNN). First, the entities and relationships in the knowledge graph need to be converted into feature vectors of nodes and edges. Each node is represented as a feature vector, and each edge needs to have a corresponding feature representation, which is generated by relationship embedding. Then, an adjacency matrix is ​​constructed to represent the connection relationship between nodes, reflecting which entities have relationships. This matrix can include different types of edge information. Finally, all node features, edge features and graph connection relationships are organized into graph structure data as the input of the graph neural network GNN. Then, the graph neural network GNN uses the information of these nodes and edges through the graph convolution mechanism to learn the deep relationship between entities, thereby realizing knowledge reasoning. After the retrieval of the chronic disease medical text data and the prompt word optimization step of the intelligent agent, the data input of the disease to be diagnosed for knowledge reasoning is finally passed to the medical large model GLM-4Local Disease for processing, thereby generating a comprehensive and accurate text output for the user. The specific network structure diagram and the large model health assistant online system operation result diagram are shown as follows. Figure 5 Shown and Figure 6 shown.

[0087] In general, in order to enhance the ability of contextual knowledge modeling, the embodiment of the present invention combines the Wikidata knowledge graph, and helps the traditional RAG model to better understand concepts, entities and their relationships through an auxiliary retrieval mechanism. It further proposes to combine the knowledge graph with the graph neural network (GNN), and use the knowledge graph embedding (KGE) technology to improve the effect of the retrieval and generation process. Through the innovative method of RAG-KGE, the model can not only enhance the understanding of complex knowledge, but also accurately quote and integrate professional knowledge in related fields during the generation process, further optimizing the quality and accuracy of the generation. The system can provide patients with accurate self-health assessments and effective disease solutions without relying on traditional medical institutions, achieving the following beneficial technical effects:

[0088] (1) Enhance self-health management capabilities: Traditional disease diagnosis and treatment usually rely on face-to-face diagnosis and doctor's diagnosis, but in a home environment, patients often cannot get professional medical services in a timely manner. The system of the present invention uses an AI-based intelligent consultation assistant, through natural language processing and dialogue generation technology, to help patients achieve health self-assessment at home. Patients only need to talk to the health assistant, and the system can provide targeted suggestions based on the huge amount of medical text data it has built, solving the problem of patients being unable to get timely medical services.

[0089] (2) Overcoming the challenge of insufficient disease data sources: Currently, many smart health applications rely on external disease data sources, which are often incomplete and have delayed updates, seriously affecting the accuracy and reliability of system diagnosis. This invention independently constructs comprehensive chronic disease medical text data (including 340,000 data items) covering fields such as internal medicine, surgery, and pediatrics, ensuring the system's comprehensive understanding and response capabilities to diseases, avoiding the limitations of insufficient or incomplete external data sources, and thus ensuring that the health assistant's diagnosis and treatment recommendations are more accurate and comprehensive.

[0090] (3) Fast and efficient diagnosis and treatment plan provision: In the traditional medical treatment model, patients need to go to the hospital to queue up, register and undergo examinations, which is a cumbersome and time-consuming process. Through the health assistant system of the present invention, patients only need to interact with the system at home through their mobile phones to quickly obtain diagnosis and treatment recommendations based on big data analysis and model reasoning, greatly improving the efficiency of diagnosis and treatment. Especially in terms of daily health management and chronic disease control, the system can provide personalized health advice and treatment plans in a timely manner to help patients better manage their own health.

[0091] (4) Personalized and preventative health management: Because the system of the present invention can comprehensively assess the patient's health status and provide personalized recommendations, patients can adjust their daily lives and diet according to the plans provided by the system, thereby achieving more efficient disease prevention and management. Chronic disease patients can slow down the progression of their disease or reduce the occurrence of complications by following the dietary habits and exercise plans provided by the system.

[0092] (5) Relieve the pressure on medical resources: With the aging of the population and the increase in the number of patients with chronic diseases, medical resources are becoming increasingly tight. The health assistant system of the present invention can alleviate this problem to a certain extent. By implementing intelligent health management in a home environment, patients can receive timely health guidance and disease warnings without occupying hospital resources, thereby providing assistance for the rational allocation of medical resources. Especially in remote areas or areas with scarce medical resources. Through the intelligent health assistant system, patients can enjoy diagnosis and treatment services similar to those of professional doctors without going to the hospital. This convenience is particularly important for patients with chronic diseases.

[0093] In summary, by combining the AI ​​big model with self-built comprehensive chronic disease medical text data, this invention can effectively solve the problems of insufficient data, untimely diagnosis and treatment, and tight medical resources in the existing medical system, and provide a more efficient, convenient, and personalized home health management and disease diagnosis and treatment plan with significant social and economic benefits.

[0094] Example 2

[0095] An embodiment of the present invention further provides a digital medical AI large-model diagnosis and treatment information retrieval system for chronic diseases, the system comprising:

[0096] A first construction unit, the first construction unit is used to obtain historical chronic disease medical data and construct chronic disease medical text data;

[0097] A first obtaining unit, wherein the first obtaining unit is used to obtain a general language model GLM-4-9B-Chat;

[0098] A second obtaining unit, the second obtaining unit is used to fine-tune the general language model GLM-4-9B-Chat using the chronic disease medical text data to obtain a large medical model GLM-4Local Disease;

[0099] a third obtaining unit, configured to obtain data of the patient's disease to be diagnosed and treated;

[0100] a fourth obtaining unit, configured to optimize and retrieve prompt words for the disease data to be diagnosed and treated using an intelligent agent to obtain RAG disease graph data text;

[0101] A first conversion unit, configured to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph using a knowledge graph embedding technology;

[0102] a fifth obtaining unit, configured to input the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated for implementing knowledge reasoning;

[0103] The first forming unit is used to input the disease data to be diagnosed and treated for realizing knowledge reasoning into the medical large model GLM-4Local Disease, generate text output, and form a consultation service health analysis report.

[0104] Example 3

[0105] Based on the same inventive concept as the digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases in the aforementioned embodiment, the present invention also provides a digital medical AI large model diagnosis and treatment information retrieval device for chronic diseases, which has a computer program stored thereon. When the program is executed by the processor, it implements the steps of any method of the digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases described above.

[0106] Among them, Figure 2 In the embodiment of the present invention, a bus architecture (represented by bus 300) is shown. Bus 300 may include any number of interconnected buses and bridges. Bus 300 links various circuits together, including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also link various other circuits together, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are not described further herein. Bus interface 306 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same component, namely a transceiver, which provides a means for communicating with various other devices over a transmission medium.

[0107] The processor 302 is responsible for managing the bus 300 and general processing, while the memory 304 may be used to store data used by the processor 302 when performing operations.

[0108] Example 4

[0109] Based on the same inventive concept as the digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases in the aforementioned embodiment, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it is used to implement the steps of the aforementioned digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases.

[0110] The above one or more technical solutions in the embodiments of the present application have at least one or more of the following technical effects:

[0111] An embodiment of the present invention provides a digital medical AI large-model diagnosis and treatment information retrieval method and system for chronic diseases, the method comprising: obtaining historical chronic disease medical data and constructing chronic disease medical text data; obtaining a general language model GLM-4-9B-Chat; using the chronic disease medical text data to fine-tune the general language model GLM-4-9B-Chat to obtain a medical large model GLM-4Local Disease; obtaining the patient's disease data to be diagnosed and treated; using an intelligent agent to optimize and retrieve prompt words for the disease data to be diagnosed and treated to obtain RAG disease graph data text; using knowledge graph embedding technology to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph; inputting the disease data to be diagnosed and treated in the form of a knowledge graph into a graph neural network to obtain disease data to be diagnosed and treated that implements knowledge reasoning; inputting the disease data to be diagnosed and treated that implements knowledge reasoning into the medical large model GLM-4Local Disease to generate text output and obtain a health analysis report for a medical consultation service. It is used to solve the problem that the medical NLP models in the existing technology generally rely on the large language model interface provided by OpenAI. Their diagnosis and treatment recommendations are usually limited to specific diseases or symptoms. When facing chronic disease consultations, there are technical problems that it is difficult to carry out comprehensive analysis and it is difficult to meet the needs of complex clinical decision-making. Based on the GLM pre-training framework, combined with the generation ability of the autoregressive model and the efficient information filling ability of the blank filling mechanism, the model can flexibly capture contextual information and fill in missing parts when processing complex texts, thereby achieving the technical effects of improving retrieval efficiency and generation process, enhancing the understanding of complex knowledge, overcoming the lack of disease data sources, providing patients with fast, efficient and accurate diagnosis and treatment plans, enhancing patients' self-health management capabilities, and alleviating medical resource pressures. The technical advantages of the large model health assistant developed by the present invention based on the large model GLM4 are specifically as follows: (1) In response to the problem of insufficient medical data for chronic diseases, the present invention searches for 340,000 medical data including surgery, internal medicine and pediatrics to form Chronic-RAG to form chronic disease medical text data. (2) Using Wikidata to create Chronic-RAG as a knowledge graph, the RAG-KGE method is used to feed it to the GNN graph neural network, which uses the information of these nodes and edges to learn the deep relationship between entities and realize knowledge reasoning. (3) The present invention designs an agent intelligent body to optimize the prompt words, uses a predefined template, and allows the intelligent body to replace the variable part in the template. By designing reasonable and accurate input prompt words, the model can be guided to generate output that better meets the needs. (4) After the user's conversation with the GLM-4Local Disease large model ends, a comprehensive and accurate text output is generated for the user, and the agent is used to form a consultation service health analysis report.

[0112] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0114] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0116] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases, characterized by: The method comprises: Obtain historical chronic disease medical data and construct chronic disease medical text data; Get the general language model GLM-4-9B-Chat; Fine-tune the general language model GLM-4-9B-Chat using the chronic disease medical text data to obtain the large medical model GLM-4Local Disease; Obtaining data on the patient's disease to be diagnosed and treated; Utilizing an intelligent agent to optimize and retrieve prompt words for the disease data to be diagnosed and treated, and obtaining RAG disease graph data text; Using knowledge graph embedding technology, the RAG disease graph data text is converted into disease data to be diagnosed and treated in the form of knowledge graph; Inputting the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated that implements knowledge reasoning; The disease data to be diagnosed and treated for implementing knowledge reasoning is input into the medical large model GLM-4Local Disease, a text output is generated, and a health analysis report of the medical consultation service is obtained.

2. The digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases according to claim 1 is characterized in that: The historical chronic disease medical data includes nearly 340,000 question and answer data in the fields of internal medicine, surgery, and pediatrics.

3. The digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases according to claim 2 is characterized in that: The process of obtaining historical chronic disease medical data and constructing chronic disease medical text data includes: Obtain historical chronic disease medical data; Classify and annotate the historical chronic disease medical data according to the chronic disease type, and associate the patient's symptoms, treatment plan, and specific medication to obtain annotated historical chronic disease medical data, wherein the chronic disease type is divided into surgery, internal medicine, and pediatrics; The annotated historical chronic disease medical data is cleaned and preprocessed to obtain chronic disease medical text data.

4. The digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases according to claim 3 is characterized in that: The cleaning and preprocessing of the annotated historical chronic disease medical data to obtain chronic disease medical text data includes: Eliminating duplicate records in the historical chronic disease medical data to obtain first historical chronic disease medical data; Performing missing value filling processing on the first historical chronic disease medical data to obtain second historical chronic disease medical data; Deleting irrelevant text information in the second chronic disease medical data to obtain third historical chronic disease medical data; Performing format unification processing on the third historical chronic disease medical data to obtain fourth historical chronic disease medical data, wherein the fourth historical chronic disease medical data meets the input data requirements of the general language model GLM-4-9B-Chat; Chronic disease medical text data is constructed based on the fourth historical chronic disease medical data.

5. The digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases according to claim 1 is characterized in that: The intelligent agent is used to optimize and retrieve prompt words for the disease data to be diagnosed and treated, and obtain RAG disease graph data text, including: Based on contextual knowledge memory query, determine whether the disease data to be diagnosed and treated has duplication issues; If there is any duplication in the disease data to be diagnosed and treated, it is directly output to the patient; According to the disease data to be diagnosed and treated, an interface function is set, wherein the interface function includes age, gender, key knowledge, and problem optimization interface function; According to the interface function, the intelligent agent is used to quickly analyze the disease data to be diagnosed and treated and extract key information; Setting priorities based on the key information; According to the priority, different tool interfaces are called to optimize and process the data of the disease to be diagnosed and treated to obtain RAG disease graph data text.

6. The digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases according to claim 1 is characterized in that: The knowledge graph embedding technology is used to convert the RAG disease graph data text into the disease data to be diagnosed and treated in the form of a knowledge graph, including: The RAG disease graph data text is made into a knowledge graph using the Wikidata data storage structure. The structural triple relationship of the knowledge graph can be expressed as (A, is_friend_of, B), where A and B are entities and "is_friend_of" is a relationship; Converting entities and relationships in the knowledge graph into feature vectors of nodes and edges, where each node is represented as a feature vector and each edge requires a corresponding feature representation generated by embedding the relationship; Constructing an adjacency matrix, wherein the adjacency matrix represents the connection relationship between nodes and reflects the existence relationship between entities; Obtaining graph structure data according to the node features, the edge features, and the connection relationship; According to the graph structure data, data on diseases to be diagnosed and treated are obtained in the form of a knowledge graph.

7. The digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases according to claim 6, characterized in that: The step of inputting the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated for implementing knowledge reasoning includes: Inputting the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network; Based on the graph convolution mechanism, the graph neural network uses the node features and the edge features to learn the deep relationships between the entities; According to the deep-level relationships between the entities, the data of the disease to be diagnosed and treated for realizing knowledge reasoning is obtained.

8. A digital medical AI large model diagnosis and treatment information retrieval system for chronic diseases, characterized by: The system comprises: A first construction unit, the first construction unit is used to obtain historical chronic disease medical data and construct chronic disease medical text data; A first obtaining unit, wherein the first obtaining unit is used to obtain a general language model GLM-4-9B-Chat; A second obtaining unit, the second obtaining unit is used to fine-tune the general language model GLM-4-9B-Chat using the chronic disease medical text data to obtain a large medical model GLM-4LocalDisease; a third obtaining unit, configured to obtain data of the patient's disease to be diagnosed and treated; a fourth obtaining unit, configured to optimize and retrieve prompt words for the disease data to be diagnosed and treated using an intelligent agent to obtain RAG disease graph data text; A first conversion unit, configured to convert the RAG disease graph data text into disease data to be diagnosed and treated in the form of a knowledge graph using a knowledge graph embedding technology; a fifth obtaining unit, configured to input the disease data to be diagnosed and treated in the form of the knowledge graph into the graph neural network to obtain the disease data to be diagnosed and treated for implementing knowledge reasoning; The first forming unit is used to input the disease data to be diagnosed and treated for realizing knowledge reasoning into the medical large model GLM-4Local Disease, generate text output, and form a consultation service health analysis report.

9. A digital medical AI large-scale model diagnosis and treatment information retrieval device for chronic diseases, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by the processor, the processor executes the steps of the digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases as described in any one of claims 1 to 7 above.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it is used to implement the steps of the digital medical AI large model diagnosis and treatment information retrieval method for chronic diseases as described in any one of claims 1 to 7 above.

Images