Intelligent generation method for traditional Chinese medicine diagnosis and treatment scheme based on knowledge graph

Abstract

The invention provides a knowledge graph-based traditional Chinese medicine diagnosis and treatment scheme intelligent generation method, and aims to provide a personalized traditional Chinese medicine treatment scheme in combination with modern medicine precise diagnosis and traditional Chinese medicine theories. A traditional Chinese medicine prescription generation model is designed by constructing a spleen and stomach system disease pathology and pharmacology knowledge graph and extracting standardized training data by using an RAG technology. The model is based on a BERT large model sequence generation technology, key information can be extracted from western medicine indexes, and a prescription conforming to the theory of traditional Chinese medicine is generated. Through model verification and optimization, the accuracy and practicability of prescription generation are improved, and clinicians can quickly obtain recommended traditional Chinese medicine prescriptions. The invention effectively promotes the modernization process of traditional Chinese medicine, and provides a new thought for improving the clinical treatment efficiency and curative effect.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence medicine. The object it processes is disease characteristics based on Western medicine diagnostic indicators, and uses retrieval enhancement to generate RAG and large model technology to achieve personalized generation of traditional Chinese medicine prescriptions. Background Art

[0002] The application of traditional Chinese medicine in modern medicine has gradually attracted attention. Diseases of the spleen and stomach system are common and frequently occurring diseases in clinical practice, and their causes are complex. Western medicine has advantages in accurate diagnosis, while Chinese medicine plays a unique role in personalized treatment. Combining the advantages of both, through RAG retrieval technology based on spleen and stomach pathology and pharmacology knowledge graph and Chinese medicine prescription generation technology based on large models, personalized and intelligent generation of Chinese medicine prescriptions can be achieved, which can effectively improve the efficiency and efficacy of clinical treatment. This research can not only promote the modernization of Chinese medicine, but also provide new ideas for global health issues. Summary of the invention

[0003] An intelligent method for generating Chinese medicine prescriptions based on characteristic indicators of Western medicine diseases. By combining the precise diagnosis of modern medicine with the theory of traditional Chinese medicine, an interdisciplinary intelligent platform is established to realize the automatic generation of Chinese medicine prescriptions, thereby providing patients with more personalized and effective treatment plans.

[0004] In order to achieve the above object, a method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph is provided, the method comprising the following steps:

[0005] Step 1: Construction of pathology and pharmacology knowledge graph

[0006] Organize the disease pathology knowledge base, construct the disease pathology knowledge map through data preprocessing and normalization, organize and analyze the pathological knowledge of systemic diseases, and build a structured knowledge base; organize the knowledge base of traditional Chinese medicine characteristics, and construct the knowledge map of traditional Chinese medicine characteristics through data preprocessing and normalization;

[0007] Step 2: Build a standard training dataset for large model training and fine-tuning

[0008] RAG technology is used to extract and organize standardized training data, clarify the input symptoms and pathological and pharmacological contexts, and define the output format of prescriptions; with the help of RAG technology, standard training data for training the large model for generating traditional Chinese medicine prescriptions is constructed, including input symptoms S1, pathological context C1, pharmacological context C2, and expected output prescription S2;

[0009] Step 3: Design and training of TCM prescription generation model

[0010] Using BERT large model sequence generation technology, we build a TCM prescription generation model applicable to various systemic diseases; we design and train a TCM prescription generation algorithm to enable it to extract key information from Western medicine indicators, and through training and iterative optimization, we enhance the model's ability to extract Western medicine diagnostic indicators and generate prescriptions that conform to TCM theory, and generate herbal formulas that conform to TCM theory;

[0011] Step 4: Model Validation and Optimization

[0012] Evaluate the effectiveness and safety of generated prescriptions in clinical applications, and based on the verification results, continuously optimize algorithms and models to improve the accuracy and practicality of prescription generation, so that clinicians can easily input patients' Western medical diagnostic indicators and quickly obtain recommended Chinese medicine prescriptions.

[0013] 2. The method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph as claimed in claim 1, wherein the specific implementation steps of step 1 are:

[0014] 1.1 Clean and standardize the content of the basic knowledge base, identify and extract key entities and relationships, such as disease names, symptoms, clinical characteristics, and drugs; use the BiLSTM-CRF algorithm for entity recognition and extract diseases, symptoms, drugs, etc. as nodes in the knowledge graph.

[0015] 1.2 Use relationship extraction technology to identify the relationship between entities and construct the edges of the knowledge graph;

[0016] 1.3 Use graph database to store and manage knowledge graphs, design graph structures to make them scalable and efficient; combine multimodal information such as images and genomic data to enrich the content of knowledge graphs, apply logical reasoning and machine learning methods to verify and derive new knowledge, and improve the accuracy and completeness of knowledge graphs; build a comprehensive, well-structured, and widely used pathology and pharmacology knowledge graph;

[0017] 1.4 For drug construction, a knowledge graph is constructed, where each medicinal material is treated as an entity, and its properties, taste, meridians and efficacy are used as attributes, and connections are established based on their combination and functional associations.

[0018] 3. The intelligent generation method of TCM diagnosis and treatment plan based on knowledge graph as claimed in claim 1, wherein the specific implementation steps of step 2 are:

[0019] 2.1 Using RAG-based enhancement technology, the pathological and pharmacological information of the disease is integrated and applied. The training samples include four key components: S1, C1, C2, and S2;

[0020] 2.2 Constructing training data set S1: Western medicine disease symptom index characteristics: Collect and organize Western medicine diagnosis data of spleen and stomach diseases, including clinical symptoms, laboratory test results, and imaging data; standardize the data, extract representative symptom index characteristics, and form the S1 data set;

[0021] 2.3 Construction of training dataset C1: Context based on RAG-enhanced disease pathology knowledge graph: Retrieve pathology information related to S1 from the constructed pathology knowledge graph using RAG technology; Generate rich contextual information, including the etiology and pathophysiological mechanism of the disease, to form the C1 dataset;

[0022] 2.4 Construction of training data set C2: Context of spleen and stomach Chinese medicine pharmacology knowledge graph enhanced by RAG: From the pharmacology knowledge graph, use RAG technology to retrieve Chinese medicine pharmacology information related to S1, provide information such as the efficacy, mechanism of action, and indications of the drug, and form the C2 data set;

[0023] 2.5 Construct training data set S2: Output the TCM prescriptions corresponding to the symptoms: Collect TCM prescription data in clinical practice, match the corresponding S1 symptom indicators, organize the prescription composition and its compatibility principle as S2 output data.

[0024] 4. The intelligent generation method of TCM diagnosis and treatment plan based on knowledge graph as claimed in claim 1, wherein the specific implementation steps of step 3 are:

[0025] 3.1 Add position encoding and segment encoding to the training data, train a multi-level Transformer model, and construct a training method for important generation sequences;

[0026] 3.2 Data preprocessing and feature encoding: vectorize and encode S1, C1, and C2 to ensure that the model can effectively process and understand multimodal information. Use text processing technology to encode S2 to maintain the semantic integrity of the traditional Chinese medicine prescription.

[0027] 3.3 Model architecture design: Design a large model architecture that combines the Transformer structure, supports RAG-enhanced context information processing, and integrates a multi-task learning mechanism so that the model can focus on pathology, pharmacology, and prescription generation tasks at the same time;

[0028] 3.4 Model training and optimization: Adopt transfer learning strategy, use pre-trained models for fine-tuning, improve the generalization ability of the model, and optimize the performance indicators of the model through cross-validation and hyperparameter tuning;

[0029] 3.5 Model performance evaluation and iterative optimization: Evaluation indicators, use precision, recall rate, F1 value, BLEU and other indicators to evaluate the performance of the model in the generation of traditional Chinese medicine prescriptions; continuous optimization, according to the evaluation results, iteratively adjust the model structure and training parameters to improve the application effect of the model.

[0030] The beneficial effects of the present invention are: combining RAG technology, large model generation technology and TCM prescription generation, realizing intelligent conversion from Western medicine diagnostic indicators to TCM prescriptions, improving the generation efficiency and personalization level of TCM prescriptions, and promoting the modernization of TCM. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is an example diagram of constructing the spleen and stomach pathology knowledge graph of the present invention.

[0032] Figure 2 This is an example diagram of constructing the knowledge graph of spleen and stomach Chinese medicine properties of the present invention.

[0033] Figure 3 Diagram of the construction of the RAG-based Chinese medicine prescription generation dataset.

[0034] Figure 4 Structural diagram of the model for generating traditional Chinese medicine prescriptions for spleen and stomach diseases. DETAILED DESCRIPTION

[0035] The following is a detailed description of various details involved in the technical solution of the present invention in conjunction with the accompanying drawings. It should be noted that the described embodiments are only intended to facilitate the understanding of the present invention and do not have any limiting effect on it.

[0036] Western medicine and Chinese and Western medicine comparison sample description:

[0037] TCM diagnosis and treatment:

[0038] Symptoms: breast lumps and accumulations in children. Pain when pressed, thinning skin, greenish-yellow complexion; imbalance of yin and yang qi in children, bloating, vomiting, abdominal pain, and lumps.

[0039] Pathology: Yin and Yang imbalance: The imbalance of Yin and Yang may lead to poor Qi flow and accumulation. Weak spleen and stomach:

[0040] Weak spleen and stomach, stagnation of food, and then the formation of fullness. Insufficient Qi and blood: Insufficient Qi and blood leads to a dull complexion.

[0041] Fatigue.

[0042] Prescription: Zhishi, Chenpi, Houpo, Cyperus, Hawthorn, Malt, Shenqu, Amomum, Aucklandia, Areca, Citrus aurantium, Licorice. Sijunzi Decoction is used to strengthen the spleen and replenish Qi, replenish Qi and nourish blood to improve Qi and blood deficiency.

[0043] Western medicine diagnosis and treatment:

[0044] Symptoms: Routine blood test: Hemoglobin: 6g / dL; Hematocrit: 22%; Liver function test:

[0045] Total protein: 3g / dL; Albumin: 1.5g / dL; C-reactive protein: 15mg / L; Pancreatic enzyme levels: pancreatic amylase and lipase: low; Weight loss, fatigue, pale complexion.

[0046] Pathology: Malnutrition: May show lower than normal weight, reflecting the phenomenon of thinning skin. Anemia: Blood tests may show low hemoglobin levels, explaining the symptom of sallow complexion. Inflammatory response:

[0047] Elevated C-reactive protein may indicate inflammation in the body.

[0048] Prescription: digestive enzyme preparations, simethicone, acetaminophen. Iron supplements to correct anemia, nutritional support to improve malnutrition.

[0049] The present invention proposes a method for generating a Chinese medicine prescription based on Western medicine diagnostic indicators. This method can effectively overcome the difficulty of understanding the traditional Chinese medicine disease diagnosis process and the lack of effective medical evidence in pathological interpretation. It effectively utilizes detailed Western medicine blood, spleen, stomach, liver and kidney detection indicators to generate a Chinese medicine prescription. The specific implementation steps are as follows:

[0050] 1. Construct training samples

[0051] The training samples include four parts, namely, Western medicine disease indicators S1, pathology knowledge base related content C1, pharmacology knowledge related content C2, and Chinese medicine prescriptions S2 that need to be output. In the above example, S1 is "blood routine examination: hemoglobin: 6g / dL; hematocrit: 22\%; liver function test: total protein: 3g / dL; albumin: 1.5g / dL; C-reactive protein: 15mg / L; pancreatic enzyme levels: pancreatic amylase and lipase: low; weight loss, fatigue, pale complexion."; C1 is "malnutrition: may show that the weight is lower than the normal standard, reflecting the phenomenon of thinning skin. Anemia: blood tests may show low hemoglobin levels, explaining the symptoms of greenish-yellow complexion. Inflammatory response: elevated C-reactive protein may indicate inflammation in the body."; S2 is "Fructus Aurantii Immaturus, Citrus Reticulatae, Magnolia Officinalis, Cyperus Root, Hawthorn, Malt, Shenqu, Amomum Villosum, Aucklandia Sinensis, Areca Nut, Citrus Aurantii, Licorice."

[0052] 2. Construct training datasets based on a large number of pathological and pathological pharmacological knowledge graphs

[0053] Based on the cases in the existing pathology database, a training data set is constructed. Each sample contains four parts: "S1, C1, C2, S2". The four parts are separated by the symbol SEP, forming a standard large language model training data set.

[0054] 3. BERT language model training

[0055] Based on the constructed data set and a large number of samples, the BERT model is trained to establish the relationship between Western medicine disease indicators and Chinese medicine treatment pharmacies. During the training process, the existing training methods of large models are used to learn from each other, and the masked attention mechanism is used for the management of different drugs and characteristics to speed up the training process of the model.

[0056] 4. Generate personalized Chinese medicine prescriptions based on new pathological diagnostic indicators

[0057] For newly emerging outpatient pathologies, we input the trained big model based on various physical examination indicators, and retrieve pathology and pharmacology-related knowledge based on RAG, input them into the big model together, and output personalized Chinese medicine diagnosis and treatment prescriptions. Based on the treatment effect, we evaluate and improve the existing data set, and continuously improve the accuracy of the big model in generating prescriptions through model fine-tuning.

[0058] The present invention provides a method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph (for ease of description, this article takes spleen and stomach diseases as an example, but is not limited to spleen and stomach diseases), and its technical solution includes the following steps:

[0059] Step S1: Construction of the spleen and stomach pathology and pharmacology knowledge graph. Specifically, organize the spleen and stomach disease pathology knowledge base, construct the spleen and stomach disease pathology knowledge graph through data preprocessing and normalization, organize and analyze the pathological knowledge of spleen and stomach system diseases, and construct a structured knowledge base; organize the spleen and stomach Chinese medicine properties knowledge base, and construct the Chinese medicine properties knowledge graph through data preprocessing and normalization.

[0060] Step S2: Construct the standard training data set required for large-scale model training and fine-tuning, apply RAG technology to extract and organize standardized training data, clarify the input symptoms and pathological pharmacological context, and define the output format of prescriptions. With the help of RAG technology, construct standard training data for training the large-scale model for generating traditional Chinese medicine prescriptions, including input symptoms S1, spleen and stomach pathological context C1, spleen and stomach pharmacological context C2, and expected output prescription S2;

[0061] Step S3: Design and train a TCM prescription generation model. Use the BERT large model sequence generation technology to build a TCM prescription generation model suitable for spleen and stomach system diseases. Design and train a TCM prescription generation algorithm to enable it to extract key information from Western medicine indicators. Through training and iterative optimization, enhance the model's ability to extract Western medicine diagnostic indicators and generate prescriptions that conform to TCM theory, and generate herbal formulas that conform to TCM theory;

[0062] Step S4: Model verification and optimization, evaluate the effectiveness and safety of the generated prescriptions in clinical applications, and continuously optimize the algorithms and models based on the verification results to improve the accuracy and practicality of prescription generation, so that clinicians can easily input the patient's Western medical diagnostic indicators and quickly obtain recommended Chinese medicine prescriptions.

[0063] Among them, in step 1, the construction of the spleen and stomach pathology knowledge graph and the pharmacology knowledge graph is realized, and the specific steps are:

[0064] 1.1 Clean and standardize the content of the basic knowledge base, identify and extract key entities and relationships, such as disease names, symptoms, clinical characteristics, drugs, etc. Use the BiLSTM-CRF algorithm for entity recognition and extract diseases, symptoms, drugs, etc. as nodes in the knowledge graph.

[0065] 1.2 Use relationship extraction techniques (such as relationship classification and coreference resolution) to identify the relationships between entities and construct the edges of the knowledge graph.

[0066] 1.3 Use graph databases (such as Neo4j) to store and manage knowledge graphs, and design graph structures to ensure good scalability and query efficiency. Combine multimodal information such as images and genomic data to enrich the content of knowledge graphs, apply logical reasoning and machine learning methods to verify and derive new knowledge, and improve the accuracy and completeness of knowledge graphs. Figure 1 As shown in the figure, the red oval represents the disease entity, and the rectangular box describes the main clinical medical characteristics of the disease. Constructing a comprehensive, well-structured, and widely used knowledge map of spleen and stomach pathology and pharmacology will provide a solid knowledge foundation for the research and clinical application of spleen and stomach diseases in combination with traditional Chinese and Western medicine.

[0067] 1.4 For the drugs of spleen and stomach system, we construct a knowledge graph, taking each medicinal material as an entity and taking its properties, taste, meridians and efficacy as attributes, such as Figure 2 As shown, connections are made based on their combination and functional associations.

[0068] In step 2, the training data set is constructed based on RAG (Retrieval-Augmented Generation):

[0069] 2.1 Using RAG-based enhancement technology, the pathological and pharmacological information of spleen and stomach diseases is integrated and applied. The training samples include four key components: S1, C1, C2, and S2.

[0070] 2.2 Construction of training data set: S1: Western medicine disease symptom index characteristics: Collect and organize Western medicine diagnosis data of spleen and stomach diseases, including clinical symptoms, laboratory test results, imaging data, etc. Standardize the data, extract representative symptom index characteristics, and form the S1 data set.

[0071] 2.3C1: Pathological knowledge graph of spleen and stomach diseases enhanced by RAG Context: From the constructed pathological knowledge graph, the RAG technology is used to retrieve pathological information related to S1. Rich contextual information is generated, including the etiology, pathophysiological mechanism, etc. of the disease, forming the C1 dataset.

[0072] 2.4C2: Context of the spleen and stomach Chinese medicine pharmacology knowledge graph based on RAG enhancement: From the pharmacology knowledge graph, use RAG technology to retrieve Chinese medicine pharmacology information related to S1. Provide information such as the efficacy, mechanism of action, and indications of the drug to form the C2 data set.

[0073] 2.5S2: Output of Chinese medicine prescriptions corresponding to symptoms: Collect Chinese medicine prescription data in clinical practice and match them with corresponding S1 symptom indicators. Sort out the prescription composition and its compatibility principle as S2 output data.

[0074] In step 3, the sequence model generation and training based on the large model of traditional Chinese medicine prescription is constructed:

[0075] 3.1 Add position encoding and segment encoding to the training data, train a multi-level Transformer model, and construct a training method for important generation sequences.

[0076] 3.2 Data preprocessing and feature encoding: S1, C1, and C2 are vectorized and encoded to ensure that the model can effectively process and understand multimodal information. Text processing technology is used to encode S2 to maintain the semantic integrity of the traditional Chinese medicine prescription.

[0077] 3.3 Model architecture design: Design a large model architecture that combines the Transformer structure to support RAG-enhanced context information processing. Integrate a multi-task learning mechanism so that the model can focus on pathology, pharmacology, and prescription generation tasks at the same time.

[0078] 3.4 Model training and optimization: Use transfer learning strategies and use pre-trained models (such as BERT and GPT) for fine-tuning to improve the generalization ability of the model. Optimize the performance indicators of the model through cross-validation and hyperparameter tuning.

[0079] 3.5 Model performance evaluation and iterative optimization: Evaluation indicators: Use precision, recall rate, F1 value, BLEU and other indicators to evaluate the performance of the model in the generation of traditional Chinese medicine prescriptions. Continuous optimization: According to the evaluation results, iteratively adjust the model structure and training parameters to improve the application effect of the model.

Claims

1. A method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph, characterized in that: The method comprises the following steps: Step 1: Construction of pathology and pharmacology knowledge graph Organize the disease pathology knowledge base, build a disease pathology knowledge map through data preprocessing and normalization, organize and analyze the pathology knowledge of system diseases, and build a structured knowledge base; Organize the knowledge base of Chinese medicine properties, and construct a knowledge graph of Chinese medicine properties through data preprocessing and normalization; Step 2: Build a standard training dataset for large model training and fine-tuning Apply RAG technology to extract and organize standardized training data, clarify input symptoms and pathological pharmacological context, and define prescription output format; With the help of RAG technology, standard training data is constructed to train the large model for generating traditional Chinese medicine prescriptions, including input symptoms S1, pathological context C1, pharmacological context C2, and expected output prescription S2; Step 3: Design and training of TCM prescription generation model Using BERT large model sequence generation technology, we build a TCM prescription generation model suitable for various systemic diseases; we design and train a TCM prescription generation algorithm to enable it to extract key information from Western medicine indicators, and through training and iterative optimization, we enhance the model's ability to extract Western medicine diagnostic indicators and generate prescriptions that conform to TCM theory, and generate herbal formulas that conform to TCM theory; Step 4: Model Validation and Optimization Evaluate the effectiveness and safety of generated prescriptions in clinical applications. Based on the verification results, continuously optimize algorithms and models to improve the accuracy and practicality of prescription generation, so that clinicians can easily input patients' Western medical diagnostic indicators and quickly obtain recommended Chinese medicine prescriptions.

2. The method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph as claimed in claim 1, characterized in that: Step 1 The specific implementation steps are: 1.1 Clean and standardize the content of the basic knowledge base, identify and extract key entities and relationships, such as disease names, symptoms, clinical characteristics, and drugs; Use the BiLSTM-CRF algorithm for entity recognition and extract diseases, symptoms, drugs, etc. as nodes in the knowledge graph; 1.2 Use relationship extraction technology to identify the relationship between entities and construct the edges of the knowledge graph; 1.3 Use graph databases to store and manage knowledge graphs, design graph structures to ensure good scalability and query efficiency; combine multimodal information such as images and genomic data to enrich the content of knowledge graphs, apply logical reasoning and machine learning methods to verify and derive new knowledge, and improve the accuracy and completeness of knowledge graphs; build a comprehensive, well-structured, and widely used pathology and pharmacology knowledge graph; 1.4 For drug construction, a knowledge graph is constructed, where each medicinal material is treated as an entity, and its properties, taste, meridians, and efficacy are used as attributes, and connections are established based on their combination and functional associations.

3. The method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph as claimed in claim 1, characterized in that: The specific implementation steps of step 2 are: 2.1 Using RAG-based enhancement technology, the pathological and pharmacological information of the disease is integrated and applied. The training samples include four key components: S1, C1, C2, and S2; 2.2 Construction of training data set S1: Western medicine disease symptom index characteristics: Collect and organize Western medicine diagnosis data of spleen and stomach diseases, including clinical symptoms, laboratory test results, and imaging data; standardize the data, extract representative symptom index characteristics, and form the S1 data set; 2.3 Construction of training dataset C1: Context based on RAG-enhanced disease pathology knowledge graph: Retrieve pathology information related to S1 from the constructed pathology knowledge graph using RAG technology; Generate rich contextual information, including the etiology and pathophysiological mechanism of the disease, to form the C1 dataset; 2.4 Construction of training dataset C2: Context of spleen and stomach Chinese medicine pharmacology knowledge graph based on RAG enhancement: From the pharmacology knowledge graph, use RAG technology to retrieve Chinese medicine pharmacology information related to S1, provide information such as the efficacy, mechanism of action, and indications of the drug, and form the C2 dataset; 2.5 Construct training data set S2: Output of TCM prescriptions corresponding to symptoms: Collect TCM prescription data in clinical practice, match the corresponding S1 symptom indicators, organize the prescription composition and its compatibility principle as S2 output data.

4. The method for intelligently generating a TCM diagnosis and treatment plan based on a knowledge graph as claimed in claim 1, characterized in that: The specific implementation steps of step 3 are: 3.1 Add position encoding and segment encoding to the training data, train a multi-level Transformer model, and construct a training method for important generation sequences; 3.2 Data preprocessing and feature encoding: S1, C1, and C2 are vectorized and encoded to ensure that the model can effectively process and understand multimodal information. Text processing technology is used to encode S2 to maintain the semantic integrity of the traditional Chinese medicine prescription. 3.3 Model architecture design: Design a large model architecture that combines the Transformer structure, supports RAG-enhanced context information processing, and integrates a multi-task learning mechanism so that the model can focus on pathology, pharmacology, and prescription generation tasks at the same time; 3.4 Model training and optimization: Adopt transfer learning strategy, use pre-trained models for fine-tuning, improve the generalization ability of the model, and optimize the performance indicators of the model through cross-validation and hyperparameter tuning; 3.5 Model performance evaluation and iterative optimization: Evaluation indicators, such as precision, recall, F1 value, BLEU, etc. are used to evaluate the performance of the model in generating traditional Chinese medicine prescriptions; Continuously optimize and iteratively adjust the model structure and training parameters based on the evaluation results to improve the application effect of the model.

Images