AI-based traditional Chinese medicine nursing recommendation system

Abstract

The invention discloses a traditional Chinese medicine nursing recommendation system based on AI. The method comprises the following steps: collecting multi-dimensional health data of a user; constructing a deep neural network model based on an improved Transform architecture, taking the multi-modal feature vector as input, taking a syndrome diagnosis result marked by a physician as a supervision signal, and outputting a syndrome state vector; performing multi-dimensional matching calculation on the syndrome state vector and a traditional Chinese medicine efficacy vector in a traditional Chinese medicine nursing knowledge base through an attention mechanism, and introducing an NSGA-III multi-objective optimization algorithm to generate an initial nursing recommendation scheme; and receiving user feedback data, inputting the user feedback data into the dynamic syndrome differentiation model module to generate a target syndrome state vector, and outputting a target traditional Chinese medicine nursing recommendation scheme. And accuracy and efficiency of traditional Chinese medicine syndromes are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an AI-based traditional Chinese medicine nursing recommendation system. Background Technology

[0002] Traditional Chinese medicine (TCM) nursing care plans rely heavily on the personal experience of TCM practitioners, which has significant limitations. Manual diagnosis is affected by factors such as the physician's skill level and work status, resulting in strong subjectivity, low efficiency, and difficulty in meeting the personalized care needs of a large number of users. TCM syndrome differentiation requires comprehensive analysis of multi-dimensional health information, including symptoms, tongue appearance, and pulse, and manual processing of complex data is limited, leading to low diagnostic efficiency. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing an AI-based traditional Chinese medicine nursing recommendation system.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following modules in the aforementioned AI-based traditional Chinese medicine nursing recommendation system:

[0005] The data acquisition and fusion module is used to collect multi-dimensional health data of users, including at least structured data and structured data. The multi-dimensional health data is preprocessed, standardized and feature-engineered in sequence to obtain multi-modal feature vectors.

[0006] The dynamic dialectical model module is used to build a deep neural network model based on the improved Transformer architecture. It takes the multimodal feature vector as input, the syndrome diagnosis results labeled by the physician as supervision signal, and outputs the syndrome state vector.

[0007] The TCM nursing recommendation module is used to perform multi-dimensional matching calculations between the syndrome state vector and the TCM efficacy vector in the TCM nursing knowledge base through the Attention mechanism, and introduces the NSGA-III multi-objective optimization algorithm to generate an initial nursing recommendation scheme.

[0008] The recommended scheme optimization module is used to receive user feedback data, generate a target syndrome state vector based on the user feedback data input into the dynamic syndrome differentiation model module, and output the target traditional Chinese medicine nursing recommendation scheme.

[0009] Furthermore, in the aforementioned AI-based traditional Chinese medicine nursing recommendation system, the data acquisition and fusion module includes the following units:

[0010] The multi-dimensional data acquisition unit is used to collect users' multi-dimensional health data. Structured data includes at least basic information, past medical history, family history, current symptoms, and objective physiological indicators; unstructured data includes at least dietary records, lifestyle habits, symptom complaint text, tongue image, and facial color image.

[0011] The structured data processing unit is used to fill missing values ​​in the structured data using the mean imputation method and to identify and delete outliers whose deviation from the mean exceeds 3 times the standard deviation using the Z-score method, thus obtaining the initial structured data.

[0012] The unstructured data processing unit is used to remove salt-and-pepper noise from images in unstructured data using a median filtering algorithm and normalize the data. It then uses a semantic segmentation model based on U-Net deep learning to segment the tongue and face regions to obtain the initial unstructured data.

[0013] Furthermore, in the aforementioned AI-based traditional Chinese medicine nursing recommendation system, the data acquisition and fusion module further includes the following units:

[0014] The data standardization unit is used to standardize the initial structured data using the Z-score standardization method and to standardize the initial unstructured data using the Min-Max standardization method, so as to obtain standard multidimensional health data.

[0015] The color feature calculation unit is used to calculate the mean, variance, skewness and kurtosis of each RGB channel of the image in the standard multidimensional health data, and to extract the mean and variance features of hue, saturation and lightness in the HSV color space.

[0016] The image feature acquisition unit is used to calculate contrast, entropy, energy, and correlation texture parameters using the GLCM gray-level co-occurrence matrix, and to extract image texture features using the LBP local binary mode.

[0017] The text vector acquisition unit is used to calculate the importance of words using the TF-IDF term frequency-inverse document frequency method. It trains the text using the Word2Vec model to generate 300-dimensional word vectors and converts the word vectors into text vectors through average pooling.

[0018] The feature data fusion unit is used to concatenate structured data features, image texture features, and text vectors to obtain a multimodal feature vector.

[0019] Furthermore, in the aforementioned AI-based traditional Chinese medicine nursing recommendation system, the dynamic syndrome differentiation model module includes the following units:

[0020] The network model building unit is used to build deep neural network models based on the improved Transformer architecture. It uses a multi-head self-attention module to capture the correlation between multimodal features from multiple perspectives and performs nonlinear transformation on the features output by the attention mechanism through a feedforward neural network module.

[0021] The supervisory signal construction unit is used to manually annotate user health data based on the theory of syndrome differentiation and treatment in traditional Chinese medicine, and to establish a labeling system containing 10 basic syndromes. One-hot encoding is used to convert each syndrome into a 10-dimensional binary vector to obtain the syndrome diagnosis result.

[0022] The syndrome state output unit is used to take the multimodal feature vector as input, the syndrome diagnosis result marked by the physician as a supervision signal, and output the syndrome state vector.

[0023] Furthermore, in the aforementioned AI-based traditional Chinese medicine nursing recommendation system, the dynamic syndrome differentiation model module further includes the following units:

[0024] The feature fusion unit is used to concatenate the preprocessed multimodal features during the model input stage to form a comprehensive feature vector containing the user's full-dimensional health information;

[0025] The dimension unification unit is used to map the concatenated high-dimensional feature vectors to a 512-dimensional unified feature space through a fully connected layer, enabling features from different modalities to interact in the same semantic space.

[0026] Furthermore, in the aforementioned AI-based traditional Chinese medicine (TCM) nursing recommendation system, the TCM nursing recommendation module includes the following units:

[0027] The setting unit is used to use the syndrome state vector output by the model as the query vector, and the efficacy vector of each Chinese medicine in the Chinese medicine nursing knowledge base as the key vector.

[0028] The quantification unit is used to quantify the degree of matching between the syndrome vector and the efficacy vector of all Chinese medicines by calculating the attention score between them.

[0029] Furthermore, in the aforementioned AI-based traditional Chinese medicine nursing recommendation system, the recommendation scheme optimization module includes the following units:

[0030] The matching degree calculation unit is used to calculate the total matching degree based on the attention mechanism, and prioritizes the selection of Chinese medicine combinations that match the top 20% of the syndrome state vector.

[0031] The recommended treatment plan generation unit is used to remove traditional Chinese medicines that conflict with the user's allergy history, while controlling the usage ratio, and generating an initial nursing care recommendation plan, which includes at least a traditional Chinese medicine combination plan, a medication guidance plan, and a supporting nursing care plan.

[0032] Furthermore, in a method for implementing an AI-based traditional Chinese medicine nursing recommendation system, the method includes the following steps:

[0033] Collect multi-dimensional health data from users, including at least structured data and structured data. Perform preprocessing, standardization, and feature engineering on the multi-dimensional health data in sequence to obtain multimodal feature vectors.

[0034] A deep neural network model is constructed based on the improved Transformer architecture. The multimodal feature vector is used as input, the syndrome diagnosis results labeled by the physician are used as supervision signals, and the syndrome state vector is output.

[0035] The Attention mechanism is used to perform multi-dimensional matching calculations between the syndrome state vector and the efficacy vector of traditional Chinese medicine in the traditional Chinese medicine nursing knowledge base, and the NSGA-III multi-objective optimization algorithm is introduced to generate an initial nursing recommendation scheme.

[0036] Receive user feedback data, input the user feedback data into the dynamic syndrome differentiation model module to generate a target syndrome state vector, and output a target TCM nursing recommendation plan.

[0037] Furthermore, in a method for implementing an AI-based traditional Chinese medicine nursing recommendation system, the method includes the following steps:

[0038] The syndrome state vector output by the model is used as the query vector, and the efficacy vector of each Chinese medicine in the Chinese medicine nursing knowledge base is used as the key vector.

[0039] The degree of matching between syndrome vectors and efficacy vectors of all Chinese herbal medicines is quantified by calculating the attention scores between them.

[0040] Furthermore, in a method for implementing an AI-based traditional Chinese medicine nursing recommendation system, the method includes the following steps:

[0041] Based on the total matching score calculated using the attention mechanism, the top 20% of traditional Chinese medicine combinations with the highest matching degree with the syndrome state vector are selected first.

[0042] Remove Chinese medicines that conflict with the user's allergy history, control the proportion of their use, and generate an initial nursing care recommendation plan, which should include at least a Chinese medicine combination plan, a medication guidance plan, and a supporting nursing care plan.

[0043] Its beneficial effects are as follows: 1. It can effectively capture the complex correlations between different modalities of data, overcoming the subjectivity and limitations of traditional manual diagnosis, and significantly improving the accuracy and efficiency of TCM syndrome differentiation. 2. It optimizes nursing plans from three dimensions: maximizing efficacy, minimizing side effects, and minimizing costs. This ensures that the recommended plans have good efficacy for the user's syndrome, avoids drug side effects, and balances cost and accessibility, making the generated TCM nursing plans more in line with actual clinical needs. 3. It can flexibly adapt to different environments and user needs, generating high-quality TCM nursing recommendations in different scenarios, improving the system's applicability and practicality. Attached Figure Description

[0044] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0045] Figure 1 This is a schematic diagram of the first embodiment of an AI-based traditional Chinese medicine nursing recommendation system according to the present invention;

[0046] Figure 2 This is a schematic diagram of a second embodiment of an AI-based traditional Chinese medicine nursing recommendation system according to the present invention;

[0047] Figure 3 This is a schematic diagram of a third embodiment of an AI-based traditional Chinese medicine nursing recommendation system according to the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0049] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0050] The present invention will now be described in detail with reference to the accompanying drawings, such as... Figure 1 As shown, an AI-based traditional Chinese medicine (TCM) nursing recommendation system includes the following modules:

[0051] The data acquisition and fusion module is used to collect users' multi-dimensional health data, including at least structured data and structured data. The multi-dimensional health data is preprocessed, standardized, and feature-engineered in sequence to obtain multi-modal feature vectors.

[0052] Specifically, this embodiment also includes a multi-dimensional data acquisition unit for collecting users' multi-dimensional health data. The structured data includes at least basic information, past medical history, family history, current symptoms, and objective physiological indicators; the unstructured data includes at least dietary records, lifestyle habits, symptom complaint text, tongue image, and facial color image.

[0053] The structured data processing unit is used to fill missing values ​​in the structured data using the mean imputation method and to identify and delete outliers whose deviation from the mean exceeds 3 times the standard deviation using the Z-score method, thus obtaining the initial structured data.

[0054] The unstructured data processing unit is used to remove salt-and-pepper noise from images in unstructured data using a median filtering algorithm and normalize the data. It then uses a semantic segmentation model based on U-Net deep learning to segment the tongue and face regions to obtain the initial unstructured data.

[0055] The data standardization unit is used to standardize the initial structured data using the Z-score standardization method and to standardize the initial unstructured data using the Min-Max standardization method, so as to obtain standard multidimensional health data.

[0056] The color feature calculation unit is used to calculate the mean, variance, skewness and kurtosis of each RGB channel of the image in the standard multidimensional health data, and to extract the mean and variance features of hue, saturation and lightness in the HSV color space.

[0057] The image feature acquisition unit is used to calculate contrast, entropy, energy, and correlation texture parameters using the GLCM gray-level co-occurrence matrix, and to extract image texture features using the LBP local binary mode.

[0058] The text vector acquisition unit is used to calculate the importance of words using the TF-IDF term frequency-inverse document frequency method. It trains the text using the Word2Vec model to generate 300-dimensional word vectors and converts the word vectors into text vectors through average pooling.

[0059] The feature data fusion unit is used to concatenate structured data features, image texture features, and text vectors to obtain a multimodal feature vector.

[0060] Specifically:

[0061] (a) Multi-dimensional health data collection;

[0062] Structured data;

[0063] Basic physiological data: Real-time collection of user data such as age, gender, height, weight, blood pressure (systolic and diastolic), and blood glucose (fasting and postprandial blood glucose) through smart wearable devices (smartwatches, health bracelets). The devices must meet medical-grade accuracy standards to ensure data accuracy.

[0064] Laboratory test data: In cooperation with medical institutions, we obtain users' routine blood indicators (white blood cell count, red blood cell count, hemoglobin, platelet count, etc.) and biochemical indicators (alanine aminotransferase, aspartate aminotransferase, serum creatinine, blood urea nitrogen, total cholesterol, triglycerides, etc.). The data is transmitted to the system through a standardized interface.

[0065] Unstructured data;

[0066] Traditional Chinese Medicine Tongue Image: Use professional tongue image acquisition equipment, such as a high-definition camera with a ring light source (resolution not less than 1920×1080). The acquisition environment should be kept with uniform lighting and a color temperature between 5500K and 6500K. The user should keep the tongue naturally extended so that the tongue body is fully exposed. The image is stored in RGB format.

[0067] Pulse description text: Pulse information is collected by experienced TCM doctors using a pulse diagnostic instrument and combined with manual diagnosis to record the characteristics of the pulse (pulse rate, pulse rhythm, pulse position, pulse strength, etc.) and input into the system in text form.

[0068] Symptom self-report text: Develop a user interface that provides a symptom selection list and a free input box. Users can fill in their symptoms via mobile app or computer, including the location, nature, duration, and aggravating or relieving factors of the symptoms. The system performs real-time grammar checks and provides guidance on standardization of the input.

[0069] Medical history records: These include the user's past medical diagnoses, treatment plans, and medication history. The system supports users uploading documents in formats such as PDF and Word, and uses OCR technology to identify and extract key information.

[0070] (ii) Data preprocessing;

[0071] Structured data processing;

[0072] Missing value handling: For a small number of missing data, the mean imputation method is used (filling based on the mean of people of the same age and gender); for indicators with a large proportion of missing data, the regression imputation method is used, using relevant variables to build a regression model for prediction and imputation.

[0073] Outlier detection and correction: Outliers are detected using the Z-score method (when a data point deviates from the mean by more than 3 times the standard deviation) and the IQR method (interquartile range). For obviously unreasonable outliers (blood pressure value of 0), correction is carried out by contacting the user or referring to historical data; for potentially reasonable outliers (low heart rate in athletes), they are retained and marked.

[0074] Unstructured data processing;

[0075] Image preprocessing: First, noise reduction is performed by using a median filter algorithm to remove salt and pepper noise from the image; then, normalization is performed to adjust the image size to 224×224 pixels to facilitate subsequent feature extraction; finally, tongue segmentation is performed by using a deep learning-based semantic segmentation model (U-Net) to accurately segment the tongue region and eliminate interference from the oral cavity background.

[0076] Text data preprocessing: For pulse description texts and symptom self-report texts, word segmentation tools specifically for the field of traditional Chinese medicine are used to segment the texts, retaining professional terminology; stop words and punctuation marks are removed; part-of-speech tagging and named entity recognition are performed on the texts to extract key symptom and syndrome-related words.

[0077] (iii) Data standardization;

[0078] The Z-score standardization method is used to make the processed structured data conform to a standard normal distribution with a mean of 0 and a standard deviation of 1. For the feature vectors obtained after processing unstructured data, the Min-Max standardization method is used to scale the data to the [0,1] interval.

[0079] (iv) Feature engineering;

[0080] Tongue image feature extraction;

[0081] Color features: Calculate the mean, variance, skewness and kurtosis of each RGB channel, and extract the hue, saturation, lightness mean and variance of the HSV color space, extracting a total of 30 color features.

[0082] Texture features: Gray-level co-occurrence matrix (GLCM) is used to calculate texture parameters such as contrast, entropy, energy, and correlation. The window size is set to 11×11, and the directions include 0°, 45°, 90°, and 135°. Four features are extracted in each direction, for a total of 16-dimensional texture features. At the same time, Local Binary Pattern (LBP) is used to extract texture features. An LBP operator with a radius of 2 and a neighborhood of 8 points is selected to generate an LBP histogram and extract 59-dimensional features.

[0083] Shape features: The perimeter, area, aspect ratio, roundness (4π×area / perimeter²), and rectangularity (area / area of ​​minimum bounding rectangle) of the tongue are calculated, and a total of 10-dimensional shape features are extracted.

[0084] Text data feature extraction;

[0085] Pulse description text and symptom self-report text: The importance of words is calculated using the TF-IDF (term frequency-inverse document frequency) method to generate TF-IDF feature vectors. The vector dimension is determined according to the size of the vocabulary (approximately 5000 dimensions). At the same time, the Word2Vec model is used to train the text to generate 300-dimensional word vectors. The word vectors are then converted into text vectors using average pooling or max pooling methods.

[0086] Multimodal feature fusion: Structured data features, tongue image features, and text data features are concatenated to obtain a multimodal feature vector, the dimension of which is the sum of the dimensions of each modality feature.

[0087] The dynamic dialectical model module is used to build a deep neural network model based on the improved Transformer architecture. It takes multimodal feature vectors as input, physician-annotated syndrome diagnosis results as supervision signals, and outputs syndrome state vectors.

[0088] Specifically, this embodiment also includes a network model building unit, which is used to build a deep neural network model based on the improved Transformer architecture, and use a multi-head self-attention module to capture the correlation between multimodal features from multiple angles; and to perform nonlinear transformation on the features output by the attention mechanism through a feedforward neural network module;

[0089] The supervisory signal construction unit is used to manually annotate user health data based on the theory of syndrome differentiation and treatment in traditional Chinese medicine, and to establish a labeling system containing 10 basic syndromes. One-hot encoding is used to convert each syndrome into a 10-dimensional binary vector to obtain the syndrome diagnosis result.

[0090] The syndrome state output unit is used to take the multimodal feature vector as input, the syndrome diagnosis results labeled by the physician as supervision signal, and output the syndrome state vector.

[0091] The feature fusion unit is used to concatenate the preprocessed multimodal features during the model input stage to form a comprehensive feature vector containing the user's full-dimensional health information;

[0092] The dimension unification unit is used to map the concatenated high-dimensional feature vectors to a 512-dimensional unified feature space through a fully connected layer, enabling features from different modalities to interact in the same semantic space.

[0093] Specifically:

[0094] (a) Core architecture design of the model;

[0095] The core model employs a modified Transformer encoder architecture, which excels at handling long-range dependencies in multimodal data and is particularly suitable for the association analysis of complex features in traditional Chinese medicine syndromes. The model consists of six identical stacked encoder layers, each containing two core modules:

[0096] Multi-head self-attention module: Supports the model to capture the correlation between multimodal features from multiple perspectives at the same time (the potential connection between tongue color and symptom text), avoiding the omission of information from a single perspective.

[0097] Feedforward neural network module: Performs nonlinear transformation on the features output by the attention mechanism to enhance the model's ability to fit complex patterns of TCM syndromes.

[0098] Within each layer, residual connections and layer normalization techniques are used to address the vanishing gradient problem during deep network training, ensuring stable model convergence.

[0099] (ii) Multimodal feature input processing;

[0100] Feature fusion strategy: An early fusion approach is adopted, directly concatenating preprocessed multimodal features (structured physiological data, image features, and symptom text features) at the model input stage to form a comprehensive feature vector containing the user's full-dimensional health information. For example, numerical features such as age and blood pressure are connected with tongue texture features and symptom keyword vectors in a fixed order to preserve the physical correlation between the original features.

[0101] Dimensional unification processing: The concatenated high-dimensional feature vectors are mapped to a 512-dimensional unified feature space through a fully connected layer (linear transformation layer), enabling features from different modalities to interact in the same semantic space, laying the foundation for efficient processing in subsequent Transformer layers.

[0102] (III) Methods for constructing supervisory signals;

[0103] Syndrome Labeling System: Based on the theory of syndrome differentiation and treatment in traditional Chinese medicine, senior TCM doctors manually label users' health data to establish a labeling system that includes 10 basic syndromes, covering common clinical syndromes (such as Qi deficiency syndrome, blood stasis syndrome, and damp-heat syndrome), and supports the expansion of more subdivided syndromes (such as Qi and Yin deficiency syndrome and phlegm and blood stasis syndrome) according to actual needs.

[0104] Label representation method: One-hot encoding is used to convert each syndrome into a 10-dimensional binary vector.

[0105] Meaning of the syndrome state vector output;

[0106] The model ultimately outputs a 256-dimensional syndrome state vector, which is an abstract representation of the multimodal features after being encoded by multiple Transformer layers, containing the following key information:

[0107] Core features of the syndrome: Select the features most relevant to the current syndrome through an attention mechanism;

[0108] Feature interaction relationships: capturing the synergistic effects between features of different modalities;

[0109] Severity of syndrome: The numerical distribution in the vector implies the severity of the syndrome (the weight of features related to "tidal fever" and "night sweats" is higher in the Yin deficiency syndrome vector).

[0110] This vector serves as the core input for the subsequent recommendation system, providing a quantitative basis for syndrome characteristics to ensure accurate matching of traditional Chinese medicine nursing plans.

[0111] The TCM nursing recommendation module is used to perform multi-dimensional matching calculations between the syndrome state vector and the TCM efficacy vector in the TCM nursing knowledge base through the Attention mechanism, and introduces the NSGA-III multi-objective optimization algorithm to generate an initial nursing recommendation scheme.

[0112] Specifically, this embodiment also includes:

[0113] The setting unit is used to use the syndrome state vector output by the model as the query vector, and the efficacy vector of each Chinese medicine in the Chinese medicine nursing knowledge base as the key vector.

[0114] The quantification unit is used to quantify the degree of matching between the syndrome vector and the efficacy vector of all Chinese medicines by calculating the attention score between them.

[0115] The matching degree calculation unit is used to calculate the total matching degree based on the attention mechanism, and prioritizes the selection of Chinese medicine combinations that match the top 20% of the syndrome state vector.

[0116] The recommended treatment plan generation unit is used to remove traditional Chinese medicines that conflict with the user's allergy history, while controlling the usage ratio, and generating an initial nursing care recommendation plan, which includes at least a traditional Chinese medicine combination plan, a medication guidance plan, and a supporting nursing care plan.

[0117] Specifically:

[0118] (a) Multi-dimensional matching calculation process;

[0119] The core role of attention mechanisms;

[0120] By establishing a mapping between syndrome state vectors and traditional Chinese medicine efficacy vectors through an attention mechanism, the core logic is to allow the model to focus on the efficacy of traditional Chinese medicine most relevant to the current syndrome. Specifically:

[0121] The 256-dimensional syndrome state vector output by the model is used as the "query vector" (representing the core features of the user's current health status).

[0122] In the TCM nursing knowledge base, the efficacy vector of each TCM serves as both a "key vector" (storing TCM efficacy tags) and a "value vector" (storing specific TCM information).

[0123] The degree of matching between the syndrome vector and all Chinese medicine efficacy vectors is quantified by calculating the "attention score" (the higher the score, the stronger the correlation; for example, the Qi deficiency syndrome vector will assign a higher score to the efficacy vectors of Qi-tonifying Chinese medicines such as Astragalus membranaceus and Codonopsis pilosula).

[0124] Multi-dimensional matching dimensions;

[0125] Based on the core theories of syndrome differentiation and treatment in Traditional Chinese Medicine, three key matching dimensions were designed:

[0126] Matching properties and meridians: Check whether the cold, hot, warm, cool properties, sour, bitter, sweet, pungent, and salty tastes of Chinese medicine, and its meridian tropism (lung meridian, spleen meridian) are consistent with the characteristics of yin-yang imbalance and organ dysfunction of the syndrome (for example, for lung heat syndrome, priority should be given to matching heat-clearing Chinese medicine that tropism to the lung meridian).

[0127] Efficacy and indication matching: Compare the correspondence between the core efficacy of Chinese medicine (tonifying qi, promoting blood circulation, and clearing heat) and the core pathogenesis of the syndrome (qi deficiency and blood stasis, damp-heat accumulation). For example, for blood stasis syndrome, focus on matching Chinese medicine with the function of promoting blood circulation and removing blood stasis.

[0128] Symptom and sign matching: The indications of traditional Chinese medicine (treatment of fatigue, improvement of purplish-dark tongue) are matched with the user's specific symptoms (self-reported fatigue, ecchymosis on the tongue);

[0129] (II) Key Points for Constructing a Traditional Chinese Medicine Nursing Knowledge Base;

[0130] Core content of the knowledge base;

[0131] A structured knowledge base comprising three main modules will be established and maintained jointly by TCM practitioners and data engineers:

[0132] Basic information about Chinese medicinal herbs: name, alternative names, properties and meridians, medicinal parts, and application in classic prescriptions (e.g., Angelica sinensis in Siwu Decoction);

[0133] Efficacy Quantification Labels: A standardized efficacy labeling system is established based on the textbook "Traditional Chinese Medicine" (containing 50+ basic efficacy labels, such as "tonifying qi and strengthening the spleen", "clearing heat and drying dampness", "cooling blood and stopping bleeding"). Each Chinese medicine is labeled with 3-5 core efficacy labels.

[0134] Efficacy vector construction method;

[0135] A strategy of "hierarchical encoding + weight assignment" is used to transform the efficacy of traditional Chinese medicine into a computable vector:

[0136] First layer (basic attributes): The properties and meridians are encoded using a unique heat code ("warm-sweet-spleen meridian" is encoded as a specific dimension combination);

[0137] The second layer (efficacy labels): uses 100-dimensional word vectors to represent efficacy labels (pre-trained through a corpus of traditional Chinese medicine literature to make the vector distance between "qi tonification" and traditional Chinese medicines such as "astragalus" and "codonopsis" closer).

[0138] The third layer (clinical weight): Based on the clinical experience of TCM doctors and the records in the pharmacopoeia, each efficacy label is assigned a weight value of 0-1 (for TCM medicines used for emergency revival, the "revival and rescue" label is given a high weight).

[0139] (III) Application of NSGA-III multi-objective optimization algorithm;

[0140] Analysis of the three major optimization objectives;

[0141] Three core optimization objectives were set based on clinical applicability:

[0142] Therapeutic efficacy priority: Based on the total matching score calculated by the attention mechanism, prioritize the selection of Chinese medicine combinations that match the top 20% of the syndrome state vector (to ensure that the core medicinal materials of the plan are targeted at the main syndrome).

[0143] Safety assurance goals: Establish a side effect risk assessment model to automatically exclude Chinese medicines that conflict with the user's allergy history, while controlling the proportion of bitter and cold / pungent and hot medicines used (to avoid excessive use of bitter and cold medicines that could damage Yin fluids in patients with Yin deficiency syndrome).

[0144] Cost adaptation target: comprehensively consider the price of Chinese medicine (refer to the average market price of Chinese medicinal materials), accessibility (distinguishing between authentic and common medicinal materials), and complexity of administration (reducing the combination of medicinal materials that require special decoction);

[0145] Key steps in the optimization process;

[0146] Solution space initialization: Using a single herb as the basic unit, generate an initial combination containing 3-5 Chinese herbs;

[0147] Pareto optimal solution set generation: Through iterative optimization of the algorithm, the solution with no obvious shortcomings in the three objectives of efficacy, safety and cost is retained;

[0148] Human-machine collaborative screening: The top 10 optimal solutions are submitted to TCM doctors for experience verification, and schemes that do not conform to the principles of TCM compatibility (combinations that violate the "Eighteen Incompatibilities" and "Nineteen Antagonisms") are excluded.

[0149] (iv) Structure of the initial nursing care recommendation program;

[0150] The solution generated after matching calculation and optimization contains four core modules:

[0151] Recommended combinations of traditional Chinese medicine;

[0152] List the names, properties, meridians, core effects, and roles of recommended Chinese herbs in the treatment plan ("Astragalus: sweet and slightly warm, enters the spleen and lung meridians, mainly tonifies qi and strengthens the exterior, targeting the main symptom of qi deficiency").

[0153] Medication guidance and advice;

[0154] Decoction methods: Distinguish between special processing requirements such as decocting first (for minerals like magnetite), adding later (for aromatic substances like mint), and decocting in a wrap (for pollen like cattail pollen);

[0155] When to take: It is recommended to take it on an empty stomach, after meals, or before bedtime (tonifying medicines are best taken on an empty stomach, while purgative medicines are best taken after meals).

[0156] The recommended solution optimization module receives user feedback data, generates a target syndrome state vector based on the user feedback data input into the dynamic syndrome differentiation model module, and outputs a target TCM nursing recommendation solution.

[0157] Specifically, in this embodiment:

[0158] (a) Collection of user feedback data;

[0159] User feedback data includes:

[0160] Satisfaction rating: Users provide feedback on their satisfaction with the recommended solution by rating (1-5 points) and providing a written description.

[0161] Feedback on effectiveness: Users report on symptom improvement after using the recommended treatment (degree of symptom relief, changes in duration, etc.), and standardized tools such as the Visual Analogue Scale (VAS) or the Symptom Checklist-90 (SCL-90) are used for quantitative evaluation.

[0162] Adverse reaction report: Users record adverse reactions (allergic reactions, gastrointestinal discomfort, etc.) that occur during the use of traditional Chinese medicine nursing programs, including symptoms, time of occurrence, and severity.

[0163] (ii) Generation of the target syndrome state vector;

[0164] User feedback data undergoes preprocessing, including text segmentation, sentiment analysis (for satisfaction rating texts), and standardization of quantitative indicators. The processed feedback data is then combined with the original multimodal feature vectors and input into a pre-trained deep neural network model. By fine-tuning the model parameters, the target syndrome state vector is generated.

[0165] (III) Output of the target-oriented traditional Chinese medicine nursing recommendation plan;

[0166] Based on the target syndrome state vector, the Attention mechanism is used again to match and calculate with the TCM efficacy vector in the TCM nursing knowledge base, and the NSGA-III multi-objective optimization algorithm is used to generate an optimized target TCM nursing recommendation scheme.

[0167] Its beneficial effects are as follows: 1. It can effectively capture the complex correlations between different modalities of data, overcoming the subjectivity and limitations of traditional manual diagnosis, and significantly improving the accuracy and efficiency of TCM syndrome differentiation. 2. It optimizes nursing plans from three dimensions: maximizing efficacy, minimizing side effects, and minimizing costs. This ensures that the recommended plans have good efficacy for the user's syndrome, avoids drug side effects, and balances cost and accessibility, making the generated TCM nursing plans more in line with actual clinical needs. 3. It can flexibly adapt to different environments and user needs, generating high-quality TCM nursing recommendations in different scenarios, improving the system's applicability and practicality.

[0168] Please see Figure 2 In an AI-based traditional Chinese medicine nursing recommendation system, the data acquisition and fusion module includes the following units:

[0169] The multi-dimensional data acquisition unit is used to collect users' multi-dimensional health data. Structured data includes at least basic information, past medical history, family history, current symptoms, and objective physiological indicators; unstructured data includes at least dietary records, lifestyle habits, symptom complaint text, tongue image, and facial color image.

[0170] The structured data processing unit is used to fill missing values ​​in the structured data using the mean imputation method and to identify and delete outliers whose deviation from the mean exceeds 3 times the standard deviation using the Z-score method, thus obtaining the initial structured data.

[0171] The unstructured data processing unit is used to remove salt-and-pepper noise from images in unstructured data using a median filtering algorithm and normalize the data. It then uses a semantic segmentation model based on U-Net deep learning to segment the tongue and face regions to obtain the initial unstructured data.

[0172] Please see Figure 3 In an AI-based traditional Chinese medicine nursing recommendation system, the data acquisition and fusion module includes the following units:

[0173] The data standardization unit is used to standardize the initial structured data using the Z-score standardization method and to standardize the initial unstructured data using the Min-Max standardization method, so as to obtain standard multidimensional health data.

[0174] The color feature calculation unit is used to calculate the mean, variance, skewness and kurtosis of each RGB channel of the image in the standard multidimensional health data, and to extract the mean and variance features of hue, saturation and lightness in the HSV color space.

[0175] The image feature acquisition unit is used to calculate contrast, entropy, energy, and correlation texture parameters using the GLCM gray-level co-occurrence matrix, and to extract image texture features using the LBP local binary mode.

[0176] The text vector acquisition unit is used to calculate the importance of words using the TF-IDF term frequency-inverse document frequency method. It trains the text using the Word2Vec model to generate 300-dimensional word vectors and converts the word vectors into text vectors through average pooling.

[0177] The feature data fusion unit is used to concatenate structured data features, image texture features, and text vectors to obtain a multimodal feature vector.

[0178] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based traditional Chinese medicine nursing recommendation system, characterized in that, The traditional Chinese medicine nursing recommendation system includes the following modules: The data acquisition and fusion module is used to collect multi-dimensional health data of users, including at least structured data and structured data. The multi-dimensional health data is preprocessed, standardized and feature-engineered in sequence to obtain multi-modal feature vectors. The dynamic dialectical model module is used to build a deep neural network model based on the improved Transformer architecture. It takes the multimodal feature vector as input, the syndrome diagnosis results labeled by the physician as supervision signal, and outputs the syndrome state vector. The TCM nursing recommendation module is used to perform multi-dimensional matching calculations between the syndrome state vector and the TCM efficacy vector in the TCM nursing knowledge base through the Attention mechanism, and introduces the NSGA-III multi-objective optimization algorithm to generate an initial nursing recommendation scheme. The recommended scheme optimization module is used to receive user feedback data, generate a target syndrome state vector based on the user feedback data input into the dynamic syndrome differentiation model module, and output the target traditional Chinese medicine nursing recommendation scheme.

2. The AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The data acquisition and fusion module includes the following units: The multi-dimensional data acquisition unit is used to collect users' multi-dimensional health data. Structured data includes at least basic information, past medical history, family history, current symptoms, and objective physiological indicators; unstructured data includes at least dietary records, lifestyle habits, symptom complaint text, tongue image, and facial color image. The structured data processing unit is used to fill missing values ​​in the structured data using the mean imputation method and to identify and delete outliers whose deviation from the mean exceeds 3 times the standard deviation using the Z-score method, thus obtaining the initial structured data. The unstructured data processing unit is used to remove salt-and-pepper noise from images in unstructured data using a median filtering algorithm and normalize the data. It then uses a semantic segmentation model based on U-Net deep learning to segment the tongue and face regions to obtain the initial unstructured data.

3. The AI-based traditional Chinese medicine nursing recommendation system as described in claim 2, characterized in that, The data acquisition and fusion module also includes the following units: The data standardization unit is used to standardize the initial structured data using the Z-score standardization method and to standardize the initial unstructured data using the Min-Max standardization method, so as to obtain standard multidimensional health data. The color feature calculation unit is used to calculate the mean, variance, skewness and kurtosis of each RGB channel of the image in the standard multidimensional health data, and to extract the mean and variance features of hue, saturation and lightness in the HSV color space. The image feature acquisition unit is used to calculate contrast, entropy, energy, and correlation texture parameters using the GLCM gray-level co-occurrence matrix, and to extract image texture features using the LBP local binary mode. The text vector acquisition unit is used to calculate the importance of words using the TF-IDF term frequency-inverse document frequency method. It trains the text using the Word2Vec model to generate 300-dimensional word vectors and converts the word vectors into text vectors through average pooling. The feature data fusion unit is used to concatenate structured data features, image texture features, and text vectors to obtain a multimodal feature vector.

4. The AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The dynamic dialectical model module includes the following units: The network model building unit is used to construct deep neural network models based on the improved Transformer architecture. It utilizes a multi-head self-attention module to capture the correlations between multimodal features from multiple perspectives; and performs nonlinear transformations on the features output by the attention mechanism through a feedforward neural network module. The supervisory signal construction unit is used to manually annotate user health data based on the theory of syndrome differentiation and treatment in traditional Chinese medicine, and to establish a labeling system containing 10 basic syndromes. One-hot encoding is used to convert each syndrome into a 10-dimensional binary vector to obtain the syndrome diagnosis result. The syndrome state output unit is used to take the multimodal feature vector as input, the syndrome diagnosis result marked by the physician as a supervision signal, and output the syndrome state vector.

5. The AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The dynamic dialectical model module also includes the following units: The feature fusion unit is used to concatenate the preprocessed multimodal features during the model input stage to form a comprehensive feature vector containing the user's full-dimensional health information; The dimension unification unit is used to map the concatenated high-dimensional feature vectors to a 512-dimensional unified feature space through a fully connected layer, enabling features from different modalities to interact in the same semantic space.

6. The AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The traditional Chinese medicine nursing recommendation module includes the following units: The setting unit is used to use the syndrome state vector output by the model as the query vector, and the efficacy vector of each Chinese medicine in the Chinese medicine nursing knowledge base as the key vector. The quantification unit is used to quantify the degree of matching between the syndrome vector and the efficacy vector of all Chinese medicines by calculating the attention score between them.

7. The AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The traditional Chinese medicine nursing recommendation module includes the following units: The matching degree calculation unit is used to calculate the total matching degree based on the attention mechanism, and prioritizes the selection of Chinese medicine combinations that match the top 20% of the syndrome state vector. The recommended treatment plan generation unit is used to remove traditional Chinese medicines that conflict with the user's allergy history, while controlling the usage ratio, and generating an initial nursing care recommendation plan, which includes at least a traditional Chinese medicine combination plan, a medication guidance plan, and a supporting nursing care plan.

8. A method for implementing an AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The method includes the following steps: Collect multi-dimensional health data from users, including at least structured data and structured data. Perform preprocessing, standardization, and feature engineering on the multi-dimensional health data in sequence to obtain multimodal feature vectors. A deep neural network model is constructed based on the improved Transformer architecture. The multimodal feature vector is used as input, the syndrome diagnosis results labeled by the physician are used as supervision signals, and the syndrome state vector is output. The Attention mechanism is used to perform multi-dimensional matching calculations between the syndrome state vector and the efficacy vector of traditional Chinese medicine in the traditional Chinese medicine nursing knowledge base, and the NSGA-III multi-objective optimization algorithm is introduced to generate an initial nursing recommendation scheme. Receive user feedback data, input the user feedback data into the dynamic syndrome differentiation model module to generate a target syndrome state vector, and output a target traditional Chinese medicine nursing recommendation plan.

9. A method for implementing an AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The method includes the following steps: The syndrome state vector output by the model is used as the query vector, and the efficacy vector of each Chinese medicine in the Chinese medicine nursing knowledge base is used as the key vector. The degree of matching between syndrome vectors and efficacy vectors of all Chinese herbal medicines is quantified by calculating the attention scores between them.

10. A method for implementing an AI-based traditional Chinese medicine nursing recommendation system as described in claim 1, characterized in that, The method includes the following steps: Based on the total matching score calculated using the attention mechanism, the top 20% of traditional Chinese medicine combinations with the highest matching degree with the syndrome state vector are selected first. Remove Chinese medicines that conflict with the user's allergy history, control the proportion of their use, and generate an initial nursing care recommendation plan, which should include at least a Chinese medicine combination plan, a medication guidance plan, and a supporting nursing care plan.

Images