Traditional Chinese medicine preference behavior prediction system and method for middle-aged and old stroke patients
By constructing a traditional Chinese medicine preference prediction model and using deep learning and transfer learning network models, the subjective and one-sided problem of traditional Chinese medicine treatment selection behavior in middle-aged and elderly stroke patients was solved, and scientific traditional Chinese medicine treatment preference prediction and personalized recommendation were achieved.
Patent Information
- Application Number
- CN202510686028.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-27
- Publication Date
- 2025-09-05
AI Technical Summary
In the existing technology, the method for determining the choice of traditional Chinese medicine treatment for middle-aged and elderly stroke patients is subjective and one-sided, lacks a scientific and effective prediction model, and is particularly affected by multi-dimensional factors such as regional differences, economic levels and social support.
Construct a TCM preference prediction model by collecting and preprocessing TCM treatment behavior data, using deep learning and transfer learning network models for training, generating a TCM preference prediction model, and outputting the patient's TCM treatment preference probability and recommended category.
It has achieved scientific prediction of the TCM treatment preferences of middle-aged and elderly stroke patients, maximized patient preference matching and minimized treatment costs, and output personalized preference weights and treatment method tendencies.
Smart Images
Figure CN120600336A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of predicting medical preferences of people, and in particular to a system and method for predicting the traditional Chinese medicine preference behavior of middle-aged and elderly stroke patients. Background Art
[0002] Cerebral stroke, also known as stroke, is clinically classified into two main types: ischemic and hemorrhagic. According to data from the Global Burden of Disease (GBD) study in 2019, there were 12.2 million new stroke cases worldwide, with a cumulative total of 101 million current cases, resulting in 143 million disability-adjusted life years (DALYs) lost and 6.55 million direct deaths. Of note, with the accelerated aging of the population, the burden of stroke in my country is expected to continue to rise. Current methods for assessing Traditional Chinese Medicine (TCM) treatment preferences among middle-aged and elderly stroke patients primarily rely on the following: 1. In-person questionnaires; 2. Online consultations and email follow-up. It is important to note that questionnaires, currently the most commonly used method in clinical settings, have limitations. These assessments are often subjective and biased, influenced by multiple confounding factors such as regional differences, economic status, and social support.
[0003] Therefore, the research on predicting the preference of traditional Chinese medicine treatment for middle-aged and elderly stroke patients is still a blank area, and a scientific and effective prediction model has not yet been established for this population. Summary of the Invention
[0004] The purpose of the present invention is to construct a TCM preference prediction model to predict the probability of patients' TCM treatment preference and the recommended TCM categories, and to provide a TCM preference behavior prediction system and method for middle-aged and elderly stroke patients.
[0005] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:
[0006] A method for predicting TCM preference behavior in middle-aged and elderly stroke patients comprises the following steps:
[0007] Step 1: Collect the initial TCM treatment behavior data of the source population, perform preprocessing, feature extraction and dimensionality reduction on the initial TCM treatment behavior data to obtain reduced-dimensional TCM feature data;
[0008] Step 2: Use the reduced-dimensional TCM feature data to train the deep learning network model and the transfer learning network model. After the training is completed, the models are integrated to generate a TCM preference prediction model.
[0009] Step 3: Input the patient's clinical data into the TCM preference prediction model to obtain the TCM treatment preference probability.
[0010] A prediction system for TCM preference behavior of middle-aged and elderly stroke patients, comprising:
[0011] The data collection module is used to collect the initial TCM treatment behavior data of the source population, perform preprocessing, feature extraction and dimensionality reduction on the initial TCM treatment behavior data, and obtain reduced-dimensional TCM feature data;
[0012] The model training module is used to train the deep learning network model and the transfer learning network model using the reduced-dimensional TCM feature data. After the training is completed, the model is integrated to generate a TCM preference prediction model.
[0013] The patient prediction module is used to input the patient's clinical data into the traditional Chinese medicine preference prediction model to obtain the probability of preference for traditional Chinese medicine treatment.
[0014] Compared with the prior art, the present invention has the following beneficial effects:
[0015] This paper constructs a multi-objective TCM preference prediction model, optimizing the model to maximize patient preference matching and minimize treatment costs. Constraints defined include medicinal material supply constraints, physician capacity limits, and patient time window requirements. The TCM preference prediction model inputs patient clinical data, outputting personalized preference weights for each patient, including TCM category preference probability, treatment method propensity score, and physician trust score. These weights are then mapped to dynamic multi-objective weight parameters. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.
[0017] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;
[0018] Figure 2 This is a schematic diagram of the preprocessing process according to an embodiment of the present invention;
[0019] Figure 3 This is a schematic diagram of the feature extraction process according to an embodiment of the present invention;
[0020] Figure 4 This is a schematic diagram of the feature dimensionality reduction process according to an embodiment of the present invention;
[0021] Figure 5This is a schematic diagram of particle swarm optimization according to an embodiment of the present invention;
[0022] Figure 6 This is a system structure diagram of an embodiment of the present invention;
[0023] Figure 7 This is a structural block diagram of the data acquisition module according to an embodiment of the present invention. DETAILED DESCRIPTION
[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.
[0025] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and are not to be understood as indicating or implying relative importance, or implying any actual relationship or order between these entities or operations. In addition, the terms "connected" and "connected" can refer to direct connection between elements or indirect connection via other elements.
[0026] Example 1:
[0027] The present invention is achieved through the following technical solutions: Figure 1 As shown, a method for predicting the TCM preference behavior of middle-aged and elderly stroke patients includes the following steps:
[0028] Step 1: Collect the initial TCM treatment behavior data of the source population, perform preprocessing, feature extraction and dimensionality reduction on the initial TCM treatment behavior data, and obtain reduced-dimensional TCM feature data.
[0029] Multi-dimensional data of Western medicine are collected from multiple medical institutions, including basic information of electronic medical records of the source population (such as name, gender, age, marital status, place of residence, etc.), DICOM image data (such as CT images, MRI images, etc.), NIHSS stroke scale scores, test indicators (such as blood tests, body fluid tests, biochemical tests, etc.), and social behavior data (such as medication records, health community semantic analysis, etc.).
[0030] Based on the information technology differences between primary hospitals and tertiary hospitals, a tiered data interface solution was designed. For example, primary hospitals could use web forms for data entry, while tertiary hospitals could connect to the HIS system via APIs. A unified data entry interface was designed, supporting both traditional Chinese and Western medicine fields (such as tongue diagnosis and pulse description in Traditional Chinese Medicine) and built-in logical validation (such as age range and the rationality of test indicators). This interface was deployed within the intranets of all levels of hospitals, enabling data aggregation via VPN or dedicated networks. Automated Data Collection Interface: HIS System Integration: Structured data such as patient gender and age were extracted using middleware (such as Mirth Connect) and synchronized daily to a central database. DICOM Image Processing: A DICOM server (such as Orthanc) was deployed to automatically receive image data and generate thumbnails and metadata indexes. Test Parameter Integration: Integration with the LIS system standardized parameter names (e.g., "HbA1c" was standardized to "glycated hemoglobin") to eliminate terminology differences between hospitals. Semantic Analysis of Health Communities: Natural Language Processing (NLP) tools (such as HanLP) were used to extract keywords (such as "dizziness" and "insomnia") from user posts to construct a symptom-syndrome association map. Data cleaning: Filter out advertisements and irrelevant content, and retain valid health-related text.
[0031] Through natural language processing (NLP), multi-dimensional data of Western medicine is converted into structured data. Blockchain hybrid encryption technology is used to encrypt the structured multi-dimensional data of Western medicine with the SM4 algorithm and store it in the alliance nodes of medical institutions in a distributed manner. The corresponding hospitals, medical insurance bureaus, and drug supervision bureaus each hold private keys to protect the privacy of the data.
[0032] Consortium chain construction: Choose a permissioned blockchain platform (such as Hyperledger Fabric), set up medical institutions, medical insurance bureaus, and drug administration bureaus as nodes, and configure CA certificates and private channels. Smart contract design: Define data access rules (e.g., "Medical Insurance Bureau can only decrypt the fee field") and log data operations in an immutable ledger. Distributed storage: Encrypted data is stored in shards across each institution's nodes, using the IPFS protocol for redundant backup and hashing to verify data integrity.
[0033] like Figure 2As shown, the initial TCM treatment behavior data of the source population are collected; the initial TCM treatment behavior data are cleaned to obtain cleaned TCM behavior data; the cleaned TCM behavior data are normalized to obtain normalized TCM behavior data; the normalized TCM behavior data are data enhanced to obtain enhanced TCM behavior data; long-term medication records are extracted from the enhanced TCM behavior data to obtain long-term medication data, and short-term treatment response extraction is performed on the enhanced TCM behavior data to obtain short-term treatment response data; the long-term medication data and short-term treatment response data are smoothed respectively to obtain smoothed long-term medication data and short-term treatment response data; the smoothed long-term medication data and short-term treatment response data are merged to obtain target TCM behavior data.
[0034] like Figure 3 As shown, the target TCM behavior data is one-hot vector encoded to obtain TCM coding input data represented by high-dimensional features; the TCM coding input data is processed by a first convolution layer through a first set of convolution kernels to obtain a first convolution feature vector; the first convolution feature vector is processed by a first pooling layer to obtain a first pooling feature vector; the first pooling feature vector is processed by a second convolution layer through a second set of convolution kernels to obtain a second convolution feature vector; the second convolution feature vector is processed by a second pooling layer to obtain a second pooling feature vector; the second pooling feature vector is processed by a fully connected layer to obtain a high-dimensional feature vector.
[0035] like Figure 4 As shown, principal component analysis is performed on the high-dimensional feature vector to obtain a principal component feature vector, and linear discriminant analysis is performed on the high-dimensional feature vector to obtain a linear discriminant feature vector. The principal component feature vector and the linear discriminant feature vector are merged to obtain a comprehensive feature vector. Finally, the comprehensive feature vector is normalized to obtain dimensionality-reduced traditional Chinese medicine feature data.
[0036] Step 2: Use the reduced-dimensional TCM feature data to train the deep learning network model and the transfer learning network model. After the training is completed, the models are integrated to generate a TCM preference prediction model.
[0037] A deep learning network model is trained using dimensionally reduced TCM feature data to obtain an initial deep learning network model. The initial deep learning module includes an input layer, a linear network layer, and an optimization layer. Model parameters are extracted from the initial deep learning network model to obtain source population model parameters. The source population model parameters are used to train a transfer learning network model for the target population to obtain an initial transfer learning network model corresponding to the target population.
[0038] In detail, Figure 5As shown, the model parameters of the initial deep learning network model are extracted to obtain the source population model parameters. The source population model parameters are then transferred to the transfer learning network model of the target population to obtain the target population model initial parameters. The initial model parameters are initialized using a particle swarm algorithm to initialize the position and velocity of the particle swarm to obtain the initial position and velocity of the particle swarm. Hyperparameter combinations are performed on the initial position and velocity of the particle swarm to obtain multiple hyperparameter combinations. The particle velocity is updated on the multiple hyperparameter combinations to obtain updated particle positions. The fitness function is calculated on the updated particle positions to obtain fitness function values. The individual optimal position is updated on the fitness function values to obtain updated individual optimal positions. The updated individual optimal positions are updated to obtain updated global optimal positions. The updated global optimal positions are then iterated through particle swarm optimization. The process of updating the velocity and position is repeated to obtain optimized particle positions and velocities. The optimized particle positions and velocities are tested for stopping conditions to obtain particle positions and velocities that meet the stopping conditions. The particle positions and velocities that meet the stopping conditions are then subjected to model parameter mapping to obtain the model target parameters. The model parameters of the transfer learning network model are updated using the model target parameters to obtain an initial transfer learning network model corresponding to the target population. The initial transfer learning network model includes a bidirectional threshold recurrent network and two fully connected layers.
[0039] A joint loss function is constructed based on cross entropy and a binary classification loss function. The initial deep learning network model and the initial transfer learning network model are trained and their parameters are tuned using the joint loss function to obtain a target deep learning network model and a target transfer learning network model. The target deep learning network model and the target transfer learning network model are integrated to generate a Traditional Chinese Medicine preference prediction model.
[0040] In detail, the reduced-dimensionality traditional Chinese medicine feature data is input into the initial deep learning network model for prediction to obtain a first prediction result, and the first loss function value corresponding to the first prediction result is calculated by the joint loss function; the model parameters of the initial transfer learning network model are trained according to the gradient information of the first loss function value and through the gradient descent optimization algorithm. At the same time, the first model parameters of the initial deep learning network model are fixed unchanged, and only the model parameters of the initial transfer learning network model are updated, wherein the first model parameters include: a first weight parameter, a first bias parameter and a first hyperparameter; the initial transfer learning network model is iteratively trained to obtain a target transfer learning network model.
[0041] The reduced-dimensional TCM feature data is input into the initial deep learning network model for prediction to obtain a second prediction result; the second loss function value corresponding to the second prediction result is calculated by the joint loss function; the second model parameters of the target transfer learning network model are fixed, and the initial deep learning model is trained according to the second loss function value to obtain the target deep learning network model.
[0042] Weights are assigned to the target deep learning network model and the target transfer learning network model to obtain a first initial weight value of the target deep learning network model and a second initial weight value of the target transfer learning network model; dynamic variable weight analysis is performed on the first initial weight value to obtain a first target weight value, and dynamic variable weight analysis is performed on the second initial weight value to obtain a second target weight value; weighted integration is performed on the target deep learning network model and the target transfer learning network model according to the first target weight value and the second target weight value to generate a traditional Chinese medicine preference prediction model.
[0043] Step 3: Input the patient's clinical data into the TCM preference prediction model to obtain the TCM treatment preference probability.
[0044] The TCM preference prediction model inputs a patient's clinical data (e.g., age, symptoms, and previous medication history). The model outputs the patient's probability of TCM treatment preference (e.g., an 80% probability of preferring a TCM compound), the recommended TCM category or combination, and explanatory features (e.g., "age > 65 years" and "previous use of Astragalus" are the main factors influencing preference).
[0045] like Figure 6 As shown, the present invention also proposes a TCM preference behavior prediction system for middle-aged and elderly stroke patients, comprising:
[0046] The data collection module is used to collect the initial TCM treatment behavior data of the source population, perform preprocessing, feature extraction and dimensionality reduction on the initial TCM treatment behavior data, and obtain reduced-dimensional TCM feature data;
[0047] The model training module is used to train the deep learning network model and the transfer learning network model using the reduced-dimensional TCM feature data. After the training is completed, the model is integrated to generate a TCM preference prediction model.
[0048] The patient prediction module is used to input the patient's clinical data into the traditional Chinese medicine preference prediction model to obtain the probability of preference for traditional Chinese medicine treatment.
[0049] Furthermore, if Figure 7 As shown, the data acquisition module specifically includes:
[0050] The data preprocessing unit is used to clean, normalize, enhance data, extract long-term medication records, extract short-term treatment responses, perform smoothing, and merge data on the initial TCM treatment behavior data to obtain the target TCM behavior data;
[0051] The feature extraction unit is used to perform one-hot vector encoding, convolution kernel processing, and fully connected layer processing on the target TCM behavior data to obtain a high-dimensional feature vector;
[0052] The feature dimensionality reduction unit is used to perform principal component analysis and linear discriminant analysis on high-dimensional feature vectors, and merge and standardize the obtained principal component feature vectors and linear discriminant feature vectors to obtain reduced-dimensional traditional Chinese medicine feature data.
[0053] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.
Claims
1. A method for predicting TCM preference behavior in middle-aged and elderly stroke patients, characterized by: The following steps are involved: Step 1: Collect the initial TCM treatment behavior data of the source population, perform preprocessing, feature extraction and dimensionality reduction on the initial TCM treatment behavior data to obtain reduced-dimensional TCM feature data; Step 2: Use the reduced-dimensional TCM feature data to train the deep learning network model and the transfer learning network model. After the training is completed, the models are integrated to generate a TCM preference prediction model. Step 3: Input the patient's clinical data into the TCM preference prediction model to obtain the TCM treatment preference probability.
2. The method for predicting the TCM preference behavior of middle-aged and elderly stroke patients according to claim 1, characterized in that: The step 1 specifically includes the following steps: Preprocessing the initial TCM treatment behavior data to obtain target TCM behavior data; Extract features from the target TCM behavior data to obtain high-dimensional feature vectors; Perform feature dimensionality reduction on high-dimensional feature vectors to obtain reduced-dimensional traditional Chinese medicine feature data.
3. The method for predicting the TCM preference behavior of middle-aged and elderly stroke patients according to claim 2, characterized in that: The step of preprocessing the initial TCM treatment behavior data to obtain target TCM treatment behavior data includes: Collect initial TCM treatment behavior data of the source population; Perform data cleaning on the initial TCM treatment behavior data to obtain cleaned TCM behavior data; Normalizing the cleaned TCM behavior data to obtain normalized TCM behavior data; Perform data enhancement on the normalized TCM behavior data to obtain enhanced TCM behavior data; Extracting long-term medication records from the enhanced TCM behavior data to obtain long-term medication data, and extracting short-term treatment response records from the enhanced TCM behavior data to obtain short-term treatment response data; The long-term medication data and the short-term treatment response data are smoothed respectively to obtain the smoothed long-term medication data and the short-term treatment response data; the smoothed long-term medication data and the short-term treatment response data are merged to obtain the target traditional Chinese medicine behavior data.
4. The method for predicting the TCM preference behavior of middle-aged and elderly stroke patients according to claim 2, characterized in that: The step of extracting features from the target TCM behavior data to obtain a high-dimensional feature vector includes: Perform one-hot vector encoding on the target TCM behavior data to obtain TCM coding input data represented by high-dimensional features; Performing a first convolution layer processing on the TCM coding input data through a first set of convolution kernels to obtain a first convolution feature vector; Performing a first pooling layer processing on the first convolutional feature vector to obtain a first pooled feature vector; Performing a second convolution layer processing on the first pooled feature vector using a second set of convolution kernels to obtain a second convolution feature vector; Performing a second pooling layer processing on the second convolutional feature vector to obtain a second pooled feature vector; Performing fully connected layer processing on the second pooled feature vector to obtain a high-dimensional feature vector.
5. The method for predicting the TCM preference behavior of middle-aged and elderly stroke patients according to claim 2, characterized in that: The step of performing feature dimensionality reduction on the high-dimensional feature vector to obtain reduced-dimensional traditional Chinese medicine feature data includes: Performing principal component analysis on the high-dimensional feature vector to obtain a principal component feature vector, and performing linear discriminant analysis on the high-dimensional feature vector to obtain a linear discriminant feature vector; Merging the principal component eigenvector and the linear discriminant eigenvector to obtain a comprehensive eigenvector; The comprehensive feature vector is standardized to obtain the dimension-reduced TCM feature data.
6. The method for predicting the TCM preference behavior of middle-aged and elderly stroke patients according to claim 1, characterized in that: The step 2 specifically includes the following steps: Use the reduced-dimensional TCM feature data to train the deep learning network model to obtain the initial deep learning network model; Extract model parameters of the initial deep learning network model to obtain the source population model parameters; use the source population model parameters to train the target population transfer learning network model to obtain the initial transfer learning network model corresponding to the target population; Constructing a joint loss function based on cross entropy and a binary classification loss function, and performing model training and model parameter tuning on the initial deep learning network model and the initial transfer learning network model using the joint loss function to obtain a target deep learning network model and a target transfer learning network model; The target deep learning network model and the target transfer learning network model are integrated to generate a traditional Chinese medicine preference prediction model.
7. The method for predicting TCM preference behavior of middle-aged and elderly stroke patients according to claim 6, characterized in that: The model parameters of the initial deep learning network model are extracted to obtain the source population model parameters; The steps of training the transfer learning network model of the target population using the model parameters of the source population to obtain the initial transfer learning network model corresponding to the target population include: Extract model parameters from the initial deep learning network model to obtain the model parameters of the source population, and transfer the model parameters of the source population to the transfer learning network model of the target population to obtain the initial model parameters of the target population; Performing particle swarm algorithm initialization on the initial parameters of the model, initializing the position and velocity of the particle swarm, and obtaining the initial position and velocity of the particle swarm; Perform hyperparameter combinations on the initial position and velocity of the particle swarm to obtain multiple hyperparameter combinations; Update particle velocity for multiple hyperparameter combinations to obtain updated particle positions; Perform fitness function calculation on the updated particle position to obtain the fitness function value; Update the individual optimal position of the fitness function value to obtain the updated individual optimal position; The updated individual optimal position is updated to the global optimal position to obtain the updated global optimal position, and the updated global optimal position is iterated by particle swarm optimization, and the process of updating the speed and position is repeated to obtain the optimized particle position and speed; Performing a stopping condition test on the optimized particle position and velocity to obtain the particle position and velocity that meet the stopping condition, and performing model parameter mapping on the particle position and velocity that meet the stopping condition to obtain the model target parameters; The model parameters of the transfer learning network model are updated using the model target parameters to obtain an initial transfer learning network model corresponding to the target population.
8. The method for predicting TCM preference behavior of middle-aged and elderly stroke patients according to claim 6, characterized in that: The step of constructing a joint loss function based on the cross entropy and the binary classification loss function, performing model training and model parameter tuning on the initial deep learning network model and the initial transfer learning network model using the joint loss function, and obtaining the target deep learning network model and the target transfer learning network model includes: Inputting the reduced-dimensional TCM feature data into the initial deep learning network model for prediction to obtain a first prediction result, and calculating a first loss function value corresponding to the first prediction result using the joint loss function; Performing model parameter training on the initial transfer learning network model using a gradient descent optimization algorithm according to the gradient information of the first loss function value, while fixing the first model parameters of the initial deep learning network model and only updating the model parameters of the initial transfer learning network model, wherein the first model parameters include: a first weight parameter, a first bias parameter, and a first hyperparameter; Iteratively training the initial transfer learning network model to obtain a target transfer learning network model; Inputting the reduced-dimensional TCM feature data into the initial deep learning network model for prediction to obtain a second prediction result, and calculating a second loss function value corresponding to the second prediction result using the joint loss function; The second model parameter of the target transfer learning network model is fixed, and the initial deep learning model is trained according to the second loss function value to obtain the target deep learning network model.
9. A system for predicting the TCM preference behavior of middle-aged and elderly stroke patients, used to implement the method according to any one of claims 1 to 8, characterized in that: include: The data collection module is used to collect the initial TCM treatment behavior data of the source population, perform preprocessing, feature extraction and dimensionality reduction on the initial TCM treatment behavior data, and obtain reduced-dimensional TCM feature data; The model training module is used to train the deep learning network model and the transfer learning network model using the reduced-dimensional TCM feature data. After the training is completed, the model is integrated to generate a TCM preference prediction model. The patient prediction module is used to input the patient's clinical data into the traditional Chinese medicine preference prediction model to obtain the probability of preference for traditional Chinese medicine treatment.
10. The TCM preference behavior prediction system for middle-aged and elderly stroke patients according to claim 9, characterized in that: The data acquisition module specifically includes: The data preprocessing unit is used to clean, normalize, enhance data, extract long-term medication records, extract short-term treatment responses, perform smoothing, and merge data on the initial TCM treatment behavior data to obtain the target TCM behavior data; The feature extraction unit is used to perform one-hot vector encoding, convolution kernel processing, and fully connected layer processing on the target TCM behavior data to obtain a high-dimensional feature vector; The feature dimensionality reduction unit is used to perform principal component analysis and linear discriminant analysis on high-dimensional feature vectors, and merge and standardize the obtained principal component feature vectors and linear discriminant feature vectors to obtain reduced-dimensional traditional Chinese medicine feature data.