Traditional Chinese medicine formula generation system based on artificial intelligence

Abstract

The invention discloses a traditional Chinese medicine formula generation system based on artificial intelligence, and belongs to the technical field of traditional Chinese medicine formula generation, and the system comprises a user interaction module which is used for obtaining the body data information of a user; the feature extraction module is used for performing feature extraction on the body data information of the user to obtain a feature data set; the intelligent formula engine is used for constructing a neural network and inputting the feature data set into the neural network for calculation to obtain a traditional Chinese medicine formula; and the efficacy prediction module is used for constructing a machine learning model, evaluating the traditional Chinese medicine formula based on the machine learning model, and generating an optimized traditional Chinese medicine formula. According to the artificial intelligence traditional Chinese medicine formula generation system, the whole process operation from user input to formula generation, drug effect prediction, result evaluation and user feedback can be realized, and an efficient and accurate traditional Chinese medicine formula generation service is provided for the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of Chinese medicine formula generation, and in particular relates to a Chinese medicine formula generation system based on artificial intelligence. Background Art

[0002] Traditional Chinese medicine is a traditional medical system with a long history. Its unique theoretical system and practical methods occupy an important position in the history of world medicine. As an important part of traditional Chinese medicine, Chinese medicine prescription is to combine multiple herbs according to specific theoretical principles to achieve the purpose of treating diseases.

[0003] However, traditional Chinese medicine prescriptions are often based on general treatment principles, which are difficult to meet modern medical needs with large individual differences. In addition, the efficacy of Chinese medicine prescriptions lacks systematic evaluation and verification by modern science, which limits their application in the global medical field. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a Chinese medicine prescription generation system based on artificial intelligence to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a Chinese medicine prescription generation system based on artificial intelligence, comprising:

[0006] A user interaction module, used to obtain the user's physical data information;

[0007] A feature extraction module, used to extract features from the user's physical data information to obtain a feature data set;

[0008] An intelligent prescription engine, used to construct a neural network, input the feature data set into the neural network for calculation, and obtain a Chinese medicine prescription;

[0009] The drug efficacy prediction module is used to build a machine learning model, evaluate the Chinese medicine prescription based on the machine learning model, and generate an optimized Chinese medicine prescription.

[0010] Preferably, the user interaction module includes:

[0011] Natural language processing engine, used to understand and process users’ natural language input;

[0012] A multimodal data input interface is used to obtain the user's physical data information based on the information uploaded by the user.

[0013] Preferably, the feature extraction module comprises:

[0014] A correlation analysis unit, used to correlate the user's physical data information to generate correlation information;

[0015] The constitution identification unit is used to determine the user's constitution type according to the traditional Chinese medicine constitution classification standard, and generate the feature data set through the associated information.

[0016] Preferably, it further comprises a herbal knowledge base, wherein the herbal knowledge base comprises a herbal interaction network and a herbal efficacy database;

[0017] The herbal medicine interaction network is used to store the relationships of mutual dependence, mutual use, mutual fear and mutual hatred among herbs;

[0018] The herbal efficacy database is used to store modern pharmacological research and clinical trial data of herbal medicines.

[0019] Preferably, the intelligent formula engine comprises:

[0020] A genetic algorithm module is used to construct a neural network, input the feature data set into the neural network for calculation, obtain a Chinese medicine prescription, and simulate natural selection and genetic mechanism to optimize the prescription;

[0021] A deep learning network is used to learn and simulate the prescription thinking of famous doctors.

[0022] Preferably, the drug efficacy prediction module comprises:

[0023] Drug efficacy database, used to obtain drug efficacy data and user feedback of historical formulas;

[0024] The prediction algorithm is used to construct a machine learning model, evaluate the Chinese medicine prescription based on the machine learning model, and generate an optimized Chinese medicine prescription.

[0025] Preferably, it also includes a result evaluation and optimization module for evaluating the effect of the optimized Chinese medicine formula and user response.

[0026] Preferably, it also includes an output interface for displaying the final Chinese medicine formula, the dosage of each herbal medicine, the efficacy prediction and theoretical basis to the user.

[0027] Compared with the prior art, the present invention has the following advantages and technical effects:

[0028] The artificial intelligence Chinese medicine prescription generation system of the present invention can realize the whole process operation from user input to prescription generation, efficacy prediction, result evaluation and user feedback, providing users with an efficient and accurate Chinese medicine prescription generation service. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0030] Figure 1 is a flowchart of an embodiment of the present invention;

[0031] Figure 2 This is a training flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0032] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0033] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] Embodiment 1

[0035] This embodiment provides a Chinese medicine prescription generation system based on artificial intelligence, including:

[0036] 1. User interaction module:

[0037] Develop a responsive graphical user interface (GUI) that provides multiple input methods, including text boxes, drop-down menus, multiple-select boxes, and sliders, to adapt to the operating habits of different users.

[0038] The integrated natural language processing (NLP) engine accurately understands the user's natural language input through semantic analysis and intent recognition, and automatically extracts key medical information.

[0039] Implement a multimodal data input interface to support users uploading medical records, test reports and medical images, as well as inputting condition descriptions through voice.

[0040] 2. Disease feature extraction module:

[0041] Implement the symptom analysis unit to identify and classify the user's symptoms and build a detailed disease model through intelligent matching with the TCM symptom database.

[0042] Develop a constitution recognition unit and use machine learning algorithms to analyze the user's constitution type, such as yin and yang, qi and blood deficiency and excess, to ensure that the prescription matches the constitution.

[0043] 3. Herbal Knowledge Base:

[0044] Build a dynamically updated herbal database that contains information on the nature, flavor, meridians, efficacy, dosage, interactions and contraindications of herbs to provide a scientific basis for prescription formulation.

[0045] Using text mining technology, we extract the usage patterns and empirical prescriptions of herbal medicine from ancient Chinese medical books and modern research literature to continuously enrich the content of the knowledge base.

[0046] 4. Formula rule library:

[0047] It collects the principles of TCM prescription, including the incompatibility of herbal medicine, the principle of monarch, minister, assistant and envoy, and famous prescriptions of ancient sages, to form a standardized prescription rule library.

[0048] Integrate modern research results, including clinical trial data and pharmacodynamic studies, to support the scientificity and effectiveness of the formulation.

[0049] 5. Intelligent formula engine:

[0050] Develop a machine learning-based matching algorithm to quickly screen out potentially effective herbal combinations based on disease characteristics and herbal knowledge base.

[0051] Genetic algorithms and deep learning techniques are used to simulate the thinking behind TCM prescriptions, optimize the ratios and combinations of herbs, and generate preliminary prescriptions.

[0052] 6. Drug efficacy prediction module:

[0053] Establish a drug efficacy database to collect historical prescription data, patient treatment effect feedback and modern pharmacology research data.

[0054] Statistical learning models, such as random forests or gradient boosting machines, are used to predict the therapeutic effects and potential side effects of prescriptions, providing a basis for prescription optimization.

[0055] 7. Result evaluation and optimization module:

[0056] Implement a real-world data analysis unit to analyze formulation effects and patient responses in actual applications to validate and improve the model.

[0057] Develop a feedback learning mechanism to adjust the formulation strategy based on patient feedback to achieve continuous model optimization and personalized adjustment.

[0058] 8. Output interface:

[0059] Design interactive charts and visualization tools to demonstrate the mechanisms of action, interrelationships, and expected effects of the herbs in the formula to enhance patient understanding.

[0060] Provides a personalized report generator to generate personalized reports containing treatment recommendations, drug efficacy predictions, diet conditioning and lifestyle guidance to support patients' comprehensive health management.

[0061] 9. System integration and optimization:

[0062] Implement modular system architecture to ensure the independence and collaborative work of each component, facilitating system expansion and maintenance.

[0063] Conduct system testing and quality assurance, including functional testing, performance testing, and user acceptance testing, to ensure system stability and reliability.

[0064] Specifically include:

[0065] (1): System construction and user interaction:

[0066] 1. System initialization: Start the artificial intelligence Chinese medicine prescription generation system, load the herbal knowledge base, prescription rule base and user interaction module.

[0067] The system checks database connections to ensure data integrity and accessibility.

[0068] The system loads built-in TCM theoretical knowledge and prescription principles to provide theoretical support for the generation process.

[0069] 2. User input: Through a graphical user interface (GUI), the user inputs specific health problems and treatment needs.

[0070] Users select symptoms, such as "headache" or "insomnia," and choose a constitution type, such as "qi deficiency" or "damp heat," from a drop-down menu.

[0071] Users describe their condition using natural language, and the NLP engine parses and extracts key information.

[0072] 3. Disease feature extraction: The system extracts disease features based on user input and combined with traditional Chinese medicine theory.

[0073] The system analyzes the match between the user's symptoms and the TCM symptom database to determine the most likely disease type.

[0074] The system adjusts the weights of disease characteristics based on the user's physical condition information to adapt to individual differences.

[0075] (2) Intelligent formula generation:

[0076] 1. Herb matching: The intelligent formula engine screens potentially effective herbs in the herbal knowledge base based on the extracted disease characteristics.

[0077] The engine uses machine learning algorithms, such as support vector machines (SVM), to match disease characteristics based on the properties and meridians of herbs.

[0078] The engine considers interactions between herbs and excludes herbal combinations that have adverse incompatibilities.

[0079] 2. Formula optimization: Based on the formula rule library, the intelligent formula engine optimizes candidate formulas.

[0080] The engine applies genetic algorithms to simulate the natural selection process and iteratively optimize the herbal ratios.

[0081] The engine is combined with a deep learning model to learn from the prescription experience of famous historical doctors and adjust the herbal combination to improve the efficacy.

[0082] (3) Efficacy prediction and result evaluation:

[0083] 1. Drug efficacy prediction: The system uses the drug efficacy prediction module to predict the therapeutic effect of the prescription.

[0084] The module uses a random forest algorithm to predict the therapeutic effects and side effects of a prescription based on historical data and a drug efficacy database.

[0085] The module takes into account the user's physical condition and the severity of the disease, and adjusts the prediction results to improve the prediction accuracy.

[0086] 2. Result evaluation: Systematically evaluate and predict drug efficacy and provide recommendations for formula adjustment.

[0087] The system analyzes the efficacy prediction results, compares them with the user's treatment goals, and evaluates the suitability of the prescription.

[0088] Based on the evaluation results and through a feedback learning mechanism, the system proposes prescription adjustment suggestions to optimize the treatment effect.

[0089] (4) Output and user feedback:

[0090] 1. Result output: The system displays the final Chinese medicine formula and treatment effect prediction to the user through the output interface.

[0091] The interface graphically displays the mechanisms of action and interrelationships of the herbs in the formula.

[0092] The interface provides personalized reports, including treatment recommendations, drug efficacy predictions, dietary adjustments, and lifestyle guidance.

[0093] 2. User feedback: The system collects user feedback on the generated formula for model optimization.

[0094] Users provide feedback through the interface to evaluate the effectiveness and satisfaction of the formula.

[0095] The system analyzes user feedback, adjusts prescription strategies, and achieves continuous optimization and personalized adjustment of the model.

[0096] Embodiment 2

[0097] like Figure 1-2 As shown, this embodiment provides a Chinese medicine prescription generation system based on artificial intelligence, including:

[0098] The prescription prediction algorithm of this version mainly generates prescriptions based on the contribution of each herb to a specific disease. First, by reading the data set containing the prescriptions and their corresponding disease labels and preprocessing the data, the herbs in each prescription are extracted and their corresponding disease labels are converted into vector representations. On this basis, each prescription in the data set is traversed, each herb and its corresponding disease label are extracted, and a disease contribution vector is constructed for each herb to represent the contribution of the herb to each disease. Then, the contribution vectors of all herbs are stored in a dictionary, where the key is the herb name and the value is the corresponding contribution vector. Finally, a herb-disease matrix is ​​formed, in which each row represents a herb and each column represents a disease.

[0099] When a new prescription is input, all the herbs in it are parsed out, and for each herb, its contribution vector is extracted from the herb-disease matrix, and the disease prediction vector of the prescription is accumulated. The disease prediction vectors are sorted, and several diseases with the highest contribution are selected as the prediction results of the prescription. This process predicts the efficacy of the prescription by simply accumulating the contribution, which reflects the relationship between herbs and diseases to a certain extent.

[0100] For the convenience of subsequent use and analysis, the herb-disease matrix was converted into DataFrame format and saved as a CSV file. In this way, researchers can easily access and analyze these data to further optimize and adjust the prediction model. In order to evaluate the accuracy of the prediction results, the inclusion rate between the actual disease and the predicted disease was calculated. The specific method is to convert the actual disease and the predicted disease into sets respectively, and then calculate the intersection of the two to obtain the inclusion rate. The inclusion rate can be used as an evaluation indicator to measure the prediction effect of the model.

[0101] This version predicts the efficacy of a prescription by simply adding up the contributions, which reflects the relationship between herbs and diseases to a certain extent. However, due to the simplicity of the model, it is unable to capture the complex interactions between herbs and deeper therapeutic mechanisms.

[0102] In terms of data sets, this embodiment uses a CSV file containing herbal names and their corresponding indexes to represent the drug data set, and each herb has a unique index. By mapping the herbal names to numerical values, the model can process these data efficiently. In addition, in order to indicate whether each drug can treat a specific disease, this embodiment uses a one-hot encoding matrix. Each row of the one-hot encoding corresponds to a herb, and the columns represent different diseases. A value of 1 indicates that the herb can treat the corresponding disease, and a value of 0 indicates that it cannot. This encoding form is not only concise but also easy for the model to understand and process. Data preprocessing is a key step to ensure the quality of model input data. This embodiment performs strict data cleaning, including removing duplicate and invalid data entries. At the same time, in order to ensure data consistency, this embodiment fills all drug sequences to the same length, making batch processing possible. This step is very important because inconsistent sequence lengths will affect the training effect of the model. In addition, this embodiment also standardizes the data to ensure the uniformity and accuracy of herbal and disease information when inputting the model.

[0103] In the actual operation process, this embodiment has carried out multiple optimizations to improve the model performance and prediction accuracy. The following are the specific measures taken in the model training and optimization of this embodiment:

[0104] Selection of loss function: This embodiment uses the cross-entropy loss function to measure the difference between the model prediction and the actual sequence. The cross-entropy loss function performs well in dealing with classification problems and can effectively guide the model to adjust parameters to make the prediction results more accurate.

[0105] Use of optimizer: In order to optimize the model parameters, this embodiment uses the Adam optimizer. The Adam optimizer combines the advantages of momentum and adaptive learning rate and can quickly converge to the optimal solution. In addition, this embodiment also uses a learning rate scheduler to dynamically adjust the learning rate to ensure the stability and efficiency of the model during training. The learning rate scheduler can automatically adjust the learning rate at different stages of training so that the model can better adapt to changes in data.

[0106] Preventing gradient explosion: Gradient explosion is a common problem in the training of deep learning models. To solve this problem, this embodiment introduces the gradient clipping technology. Each time the model parameters are updated, this embodiment clips the gradient and limits its maximum value, thereby preventing the model from being unstable due to excessive gradients.

[0107] Optimization of model structure: In order to improve the expressiveness and prediction performance of the model, this embodiment adjusts the number of layers of the Transformer decoder. By increasing the number of decoder layers, this embodiment can capture more complex sequence relationships. In addition, this embodiment also optimizes the embedding dimension, the number of attention heads, and the dimension of the feedforward neural network to meet the needs of specific tasks. These adjustments make the model perform better when processing large-scale data.

[0108] Adjustment of batch size: Batch size has an important impact on the training efficiency and performance of the model. According to the memory capacity of the GPU, this embodiment adjusts the batch size during training to improve the training efficiency. An appropriate batch size can not only speed up the training speed, but also improve the generalization ability of the model to a certain extent.

[0109] In the actual prediction process, this embodiment starts with a set of starting drugs and gradually predicts the drugs that may appear next through the model. Specifically, the model generates a drug probability distribution based on the current sequence and one-hot encoding information, and selects the next drug from it. This process continues until the complete drug sequence is generated or the maximum sequence length is reached.

[0110] For example, assuming that the input starting drug sequence is Angelica and Astragalus, the model will gradually generate subsequent drug sequences. At each step, the model will calculate the probability distribution of the next drug based on the current sequence and the unique hot encoding information, and then select the most likely drug from it. This prediction process not only improves the accuracy of drug combinations, but also provides new ideas and methods for Chinese medicine formulas. By analyzing and verifying the generated drug sequences, this embodiment can discover potential drug combinations, further optimize Chinese medicine formulas, and improve their therapeutic effects. For example, the drug sequences generated by the model can be verified in the laboratory, and effective drug combinations can be screened out, providing strong support for clinical applications.

[0111] First, in order to reduce the model's dependence on the order of herbs, the order of herbs in each prescription can be randomly shuffled during the data preprocessing stage, and the order can be re-shuffled regularly during the training process to ensure that the model can effectively learn in different orders. This will help improve the generalization ability of the model and avoid overfitting problems. Second, the quality and scope of the data set directly affect the training effect and prediction performance of the model: by cooperating with more TCM research institutions, obtaining more high-quality TCM formula data, expanding the scope of the data set to cover more different types of diseases and treatments, and cleaning and standardizing the existing data, more information and richer samples can be provided to the model, thereby improving the generalization ability of the model.

[0112] In terms of model architecture, we can increase the number of encoder and decoder layers on the basis of the existing Transformer model, and introduce technologies such as residual connection and layer normalization to improve the training effect of the model. These adjustments will improve the model's expressiveness and learning ability and capture more complex sequence relationships. In addition, combined with other advanced deep learning technologies, such as an improved version of the Attention mechanism and different types of activation functions, the model structure can be further optimized to improve the model's complexity and training effect. In order to more comprehensively evaluate the model performance, in addition to the accuracy rate, other evaluation indicators such as precision, recall rate and F1 score can be introduced to perform multi-dimensional analysis on the model's prediction results and evaluate the performance in different scenarios. Establishing a detailed evaluation report to show the performance of each indicator on different test sets can more comprehensively discover potential problems and room for improvement, and guide the optimization and adjustment of the model. In terms of result output, a random forest decision system composed of training sets of different evaluation indicators can be established to enhance the robustness of the system.

[0113] Data enhancement technology is also an important direction for improvement. By performing various transformations on existing data, generating new data samples, and increasing data diversity, the generalization ability of the model can be improved. For example, existing prescriptions can be randomly combined and replaced to generate new prescription samples; virtual prescriptions can be generated using synthetic data technology to expand the scope and diversity of the data set; noise and random perturbations can be introduced to enhance the robustness of the model to different data inputs. Model integration is also an effective improvement method. By combining multiple models with different architectures and training strategies, the overall prediction performance can be improved. For example, multiple models with different architectures can be trained and their prediction results can be weighted averaged or voted; the prediction results of multiple models can be fused using an ensemble learning algorithm; and multi-level integration methods can be introduced to optimize and fuse model results step by step.

[0114] The GraphSAGE model is a graph neural network model that aims to learn low-dimensional representations of nodes. Its core idea is to learn the representation of nodes by sampling neighbor nodes and aggregating their features. This method enables GraphSAGE to be extended to large-scale graphs and achieve good performance in tasks such as node classification, link prediction, and graph classification. The model was used on a large number of herbal-ingredient datasets in this example to embed the graph structure into a vector representation and obtain good results.

[0115] Neighbor sampling is the first step of GraphSAGE. Neighbor sampling means that each node randomly selects several of its neighboring nodes. A common practice is to set a parameter K, and each node selects its K neighbors. Feature aggregation refers to the aggregation of each node's own features and the features of the sampled neighboring nodes. The aggregation operation is similar to Pooling in CNN, and you can choose Mean, Max, etc. The aggregated feature vector passes through a fully connected layer to iteratively generate new features for each node.

[0116] The model used in this example learns node features through two convolutional layers. The first convolutional layer converts the node features into 128 dimensions, and the second layer further processes them into the output dimension. In the forward propagation, the node features pass through these two layers, and the ReLU activation and dropout layers are used to enhance the expressiveness of the model and prevent overfitting, thereby improving the performance of graph data tasks.

[0117] The cross loss function includes positive sample loss and negative sample loss, and the model is trained by positive sample pairs and negative sample pairs. It can effectively shorten the embedding distance of similar nodes and increase the embedding distance of dissimilar nodes, thereby better reflecting the similarity of nodes in the node embedding representation.

[0118] Assume that there are two sets P and N in this embodiment, representing the positive sample pairs and negative sample pairs in the graph respectively. Set the edge (i, j) to represent the positive sample pair and the node pair (k, l) to represent the negative sample pair. The vectors are represented as ei and ej.

[0119] Positive sample distance:

[0120]

[0121] Negative sample distance:

[0122]

[0123] The overall cross loss is the sum of the two. The loss function of this embodiment takes into account the positive and negative sample pairs, which is more complex in calculation and processing, but has better effect.

[0124] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A Chinese medicine prescription generation system based on artificial intelligence, characterized in that: include: A user interaction module, used to obtain the user's physical data information; A feature extraction module, used to extract features from the user's physical data information to obtain a feature data set; An intelligent prescription engine, used to construct a neural network, input the feature data set into the neural network for calculation, and obtain a Chinese medicine prescription; The drug efficacy prediction module is used to build a machine learning model, evaluate the Chinese medicine prescription based on the machine learning model, and generate an optimized Chinese medicine prescription.

2. The artificial intelligence-based Chinese medicine formula generation system according to claim 1, characterized in that: The user interaction module comprises: Natural language processing engine, used to understand and process users’ natural language input; A multimodal data input interface is used to obtain the user's physical data information based on the information uploaded by the user.

3. The Chinese medicine formula generation system based on artificial intelligence according to claim 1, characterized in that: The feature extraction module comprises: A correlation analysis unit, used to correlate the user's physical data information to generate correlation information; The constitution identification unit is used to determine the user's constitution type according to the traditional Chinese medicine constitution classification standard, and generate the feature data set through the associated information.

4. The artificial intelligence-based Chinese medicine formula generation system according to claim 1, characterized in that: Also included is an herbal knowledge base, the herbal knowledge base comprising an herbal interaction network and an herbal efficacy database; The herbal medicine interaction network is used to store the relationships of mutual dependence, mutual use, mutual fear and mutual hatred among herbs; The herbal efficacy database is used to store modern pharmacological research and clinical trial data of herbal medicines.

5. The artificial intelligence-based Chinese medicine formula generation system according to claim 1, characterized in that: The intelligent formula engine comprises: A genetic algorithm module is used to construct a neural network, input the feature data set into the neural network for calculation, obtain a Chinese medicine prescription, and simulate natural selection and genetic mechanism to optimize the prescription; A deep learning network is used to learn and simulate the prescription thinking of famous doctors.

6. The Chinese medicine formula generation system based on artificial intelligence according to claim 1, characterized in that: The drug efficacy prediction module comprises: Drug efficacy database, used to obtain drug efficacy data and user feedback of historical formulas; The prediction algorithm is used to construct a machine learning model, evaluate the Chinese medicine prescription based on the machine learning model, and generate an optimized Chinese medicine prescription.

7. The artificial intelligence-based Chinese medicine formula generation system according to claim 1, characterized in that: It also includes a result evaluation and optimization module for evaluating the effect of the optimized Chinese medicine formula and user response.

8. The Chinese medicine formula generation system based on artificial intelligence according to claim 1, characterized in that: It also includes an output interface for showing the user the final Chinese medicine formula, the dosage of each herb, the efficacy prediction and the theoretical basis.

Images