AI-based traditional Chinese medicine dialectical auxiliary diagnosis method and system and medium thereof

Abstract

The invention discloses an AI-based traditional Chinese medicine dialectical auxiliary diagnosis method and system and a medium thereof, and relates to the technical field of intelligent auxiliary diagnosis, and the method comprises the steps: collecting physiological monitoring data, tongue condition data and pulse condition signals of a patient, obtaining patient symptom information, mapping the patient symptom information to a preset traditional Chinese medicine pathogenesis classification model, and generating an initial symptom feature vector; calling a knowledge base containing traditional Chinese medicine prescription rules and compatibility medication taboo, performing data space-time alignment processing and dynamic combination calculation on the initial symptom feature vector, and generating a personalized candidate prescription set conforming to the traditional Chinese medicine compatibility taboo; according to dynamic combination calculation, a solution algorithm based on a constraint satisfaction problem is adopted, and multi-dimensional association rule matching is carried out in combination with the mapping relation between syndromes and modern medical symptoms; and outputting a prescription medication recommendation scheme containing traditional Chinese medicine compatibility, dosage and decoction methods based on the candidate prescription set. The method supports the formulation of an accurate personalized diagnosis and treatment scheme, and meets the requirements of high-quality traditional Chinese medicine personalized auxiliary diagnosis and treatment.

Description

Technical Field

[0001] The present application relates to the field of intelligent assisted diagnosis and treatment technology, and in particular to an AI-based Chinese medicine dialectical assisted diagnosis method, system and medium thereof. Background Art

[0002] Currently, the Traditional Chinese Medicine (TCM) industry faces challenges such as uneven distribution of medical resources, physician shortages, varying physician skills, and difficulties in accumulating practical experience. While existing technologies attempt to improve efficiency through digitalization, they lack effective intelligent tools to support the development of precise, personalized diagnosis and treatment plans.

[0003] Currently, some basic electronic medical record systems and simple symptom self-checking tools have been used in clinical practice, but they have failed to fully utilize big data analysis and artificial intelligence algorithms to provide personalized TCM prescription recommendations. They have limitations in adapting to individual differences in patients and cannot meet the needs of high-quality personalized TCM auxiliary diagnosis and treatment. Therefore, they need to be improved. Summary of the Invention

[0004] In order to support the formulation of accurate personalized diagnosis and treatment plans and meet the needs of high-quality personalized auxiliary diagnosis and treatment of traditional Chinese medicine, this application provides an AI-based traditional Chinese medicine dialectical auxiliary diagnosis method, system and medium.

[0005] In the first aspect, the invention objectives of this application are achieved by adopting the following technical solutions: AI-based TCM dialectical auxiliary diagnosis methods include: Collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information; Mapping the patient's symptom information to a preset TCM pathogenesis classification model to generate an initial symptom feature vector; The knowledge base containing TCM prescription rules and compatibility taboos is called up, and the initial symptom feature vector is subjected to data spatiotemporal alignment processing and dynamic combination inference through a deep learning model to generate a personalized candidate prescription set that meets TCM compatibility taboos. The dynamic combination inference adopts a solution algorithm based on the constraint satisfaction problem, combining the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching; Based on the candidate prescription set, a prescription medication recommendation plan including Chinese medicine compatibility, dosage and decoction method is output.

[0006] By adopting the above technical solutions, physiological monitoring data such as patient's blood pressure, heart rate and body temperature data; tongue data including tongue coating thickness, tongue color index and crack degree; pulse signals including pulse rate, pulse shape and pulse strength; patient data collection based on multimodal data, constructing a patient state vector containing multiple types of feature dimensions, realizing the digital quantification of TCM "look, listen, ask and feel" four diagnostic information; compared with traditional single tongue diagnosis or pulse diagnosis, multi-dimensional data fusion improves the completeness of patient symptom information, and provides a more comprehensive objective basis for accurate TCM syndrome differentiation and auxiliary diagnosis; this application uses dynamic combination calculation to realize personalized prescription generation, and this application combines machine learning algorithms with a knowledge base containing TCM prescription rules and compatibility taboos (such as TCM "Eighteen Antidotes" and "Nineteen Fears"), which can quickly complete big data in a short time. Through spatial retrieval of a large number of prescription combinations (such as 10^6 kinds), the optimal solution set that conforms to the TCM compatibility taboos is generated. Through the mapping relationship between syndromes and modern medical symptoms (such as mapping "bitter mouth and dry throat" to "liver and gallbladder dampness and heat"), the traditional dialectical thinking and modern medical knowledge are cross-validated. By integrating the traditional TCM knowledge base with modern AI technology, accurate TCM pathogenesis analysis is achieved, thereby obtaining a personalized candidate prescription set for individual symptoms. This application improves the effect of TCM dialectical auxiliary diagnosis by utilizing AI technology, and finally outputs an accurate prescription and medication recommendation scheme obtained by efficient data processing and analysis based on AI technology, which is conducive to assisting TCM in making the final diagnosis and treatment plan. It has significant clinical application value in application scenarios such as online medical platforms, remote diagnosis and treatment services, and primary medical institutions.

[0007] In a preferred example of the present application, the data spatiotemporal alignment processing and dynamic combination inference of the initial symptom feature vector by the deep learning model include: The physiological monitoring data is standardized to construct a time series feature vector including blood pressure, heart rate, and body temperature; the tongue coating thickness, tongue color index, and crack degree characteristic parameters of the tongue image data are extracted using an image segmentation algorithm to construct a tongue image feature matrix; Adopting short-time Fourier transform to extract the time domain characteristics and frequency domain characteristics of the pulse signal to generate a pulse characteristic vector; The dynamic time warping algorithm is used to align tongue, pulse, and physiological monitoring features in time and space to generate a patient state vector containing multiple feature dimensions. The multimodal features aligned by the multi-head attention mechanism are weightedly fused to generate a comprehensive pathogenesis probability distribution matrix.

[0008] By adopting the above technical solutions, the standardized processing of physiological monitoring data is conducive to eliminating the difference in the dimension of the winning signal, such as the unit difference between blood pressure mmHg and heart rate bpm; making the data collected by different equipment (such as home sphygmomanometers and hospital monitors) comparable; quantification of tongue coating thickness (μm-level accuracy) can distinguish between damp-heat / cold-damp syndrome; tongue color index (RGB-Lab space conversion) is more objective than traditional visual judgment; the grading of crack degree (0-3 levels) is significantly correlated with Yin deficiency / blood stasis syndrome (p<0.05); the time domain characteristics of pulse characteristics (referring to the peak interval) reflect the heart rhythm, and the frequency domain characteristics (referring to the energy distribution) are related to the rise and fall of Qi and blood. Different types of multimodal data are calculated through the multi-head attention fusion mechanism to generate decision-making with higher efficiency and accuracy, thereby reducing the entropy value of the comprehensive pathogenesis probability distribution matrix.

[0009] In a preferred example of the present application, the method further includes: calling a knowledge base containing TCM prescription rules and compatibility taboos, performing data spatiotemporal alignment processing and dynamic combination inference on the initial symptom feature vector through a deep learning model, and extracting frequently occurring TCM compatibility patterns based on a historical case database using an association rule mining algorithm; The effectiveness of the compatibility patterns is scored through machine learning models to screen out compatibility rules that meet the preset TCM pharmacopoeia standards; the screened compatibility rules are integrated with modern medical taboos to construct a dynamically updated TCM knowledge graph; Based on the comprehensive pathogenesis probability distribution matrix and the TCM knowledge graph, a personalized candidate prescription set that complies with TCM compatibility taboos is generated.

[0010] By adopting the above technical solutions, the ability to discover TCM diagnosis and treatment knowledge based on AI technology is improved. This application uses association rules to mine high-frequency TCM compatibility patterns. Based on the historical case database, it helps to discover effective treatment combinations and provide a reference basis for new cases; the machine learning model scores and screens the compatibility rules to ensure that the recommended prescriptions are both effective and safe, such as FDA blacklisted drugs (such as ephedra and monoamine oxidase inhibitors) to avoid interactions between Chinese and Western medicines; the dynamically updated TCM knowledge graph is also more adaptable and practical; this application combines the patient's comprehensive pathogenesis probability distribution matrix and the TCM knowledge graph to generate a personalized set of candidate prescriptions that meet TCM compatibility contraindications, thereby improving the pertinence and effectiveness of the treatment plan.

[0011] In a preferred embodiment of the present application, the patient's physiological monitoring data, tongue image data, and pulse signal are collected to obtain the patient's symptom information, including: Wearable devices collect patients' blood pressure, heart rate, and body temperature data in real time, perform noise filtering and fill in missing values ​​to obtain preprocessed physiological monitoring data; The patient's tongue image is acquired based on an image acquisition device, and a color correction algorithm is used to eliminate light interference to obtain standardized tongue image data; The pulse signal of the patient is collected through the pulse sensor, and the high-frequency noise is removed by wavelet transform to obtain the pulse signal.

[0012] By adopting the above technical solution, the patient's blood pressure, heart rate, and body temperature data are collected in real time through wearable devices, and noise filtering and missing value filling are performed to ensure the quality and reliability of physiological monitoring data; the patient's tongue image is obtained based on the image acquisition device, and the color correction algorithm is used to eliminate light interference to obtain standardized tongue data, ensuring the accuracy of the tongue analysis results; the pulse signal is collected through the pulse sensor, and the wavelet transform is applied to remove high-frequency noise, thereby improving the quality of the pulse signal.

[0013] In a preferred example of the present application, the dynamic combination calculation adopts a genetic algorithm-based constraint satisfaction problem solution, including: Encode the incompatibility rules of traditional Chinese medicine and generate chromosome gene sequences; Evaluating the compliance of candidate prescriptions by a fitness function, wherein the compliance includes contraindication conflict rate and drug efficacy synergy; Generate new prescriptions using crossover and mutation operations, and optimize the solution space using Monte Carlo tree search; Output the Top-K candidate prescription sets that meet the constraints.

[0014] By adopting the above technical solution, chromosome encoding refers to converting the Chinese medicine compatibility taboo rules into gene sequences, using a hybrid method of binary coding + weight coding. Gene segment 1 represents the prohibited drug pair (such as "Aconite + Pinellia"), and segment 2 represents the dosage ratio, thereby realizing the structured representation of Chinese medicine knowledge, which is convenient for mathematical modeling and optimization in the AI ​​system; the taboo conflict rate CR = the number of conflicting drug pairs / the total number of drug pairs, the pharmacodynamic synergy CS = the effective ingredient synergy index, and a dual-objective optimization method based on the taboo conflict rate CR and the pharmacodynamic synergy CS is adopted to improve the scientific nature of prescription recommendations, taking into account both safety and efficacy; the genetic algorithm is combined with the Monte Carlo tree to simulate real patient feedback, so as to quickly find the optimal solution in large-scale prescription combinations, significantly improve the search efficiency and global convergence ability, and is suitable for complex and nonlinear prescription combination problems.

[0015] In a preferred example of this application, the dynamic update mechanism of the TCM knowledge graph includes: Collect patient medication feedback data and construct adverse reaction knowledge triples based on drugs, symptoms, and weight coefficients; The knowledge graph embedding algorithm is used to update the entity relationship vector and dynamically adjust the credibility of the incompatibility rules; When the confidence level of a certain incompatibility is lower than the preset confidence threshold, the expert review process is triggered for manual verification; Based on federated learning, the updated TCM clinical data is integrated and the modern medicine contraindications module of the TCM knowledge graph is incrementally updated.

[0016] By adopting the above technical solution, the triple structure (drug-symptom-weight) is used to represent adverse reactions, thereby realizing the effective use of actual clinical data; then the embedding model is applied to continuously update the relationship vector, thereby improving the intelligence and adaptability of the knowledge graph so that it can reflect the latest clinical evidence; then, the confidence threshold and manual review mechanism are set to balance the automated update and manual control to ensure the reliability and security of the knowledge graph update; this application integrates the updated clinical data of traditional Chinese medicine based on federated learning, supports cross-institutional knowledge sharing, and improves the generalization ability of the system.

[0017] In a preferred example of the present application, the knowledge graph embedding algorithm is the TransE algorithm; the preset confidence threshold is a dynamic threshold; and the dynamic update mechanism of the TCM knowledge graph further includes: Based on the BERT model, semantic embedding of adverse reaction knowledge triples is performed to calculate the contextual similarity of the drug-symptom relationship. A dynamic threshold τ = α × std(C) + β × mean(C) is set, where α and β are adjustment coefficients. When the confidence level C of a certain incompatibility is less than τ, an early warning is triggered. The gradient update in federated learning is perturbed through the differential privacy mechanism, and the perturbation amplitude Among them, δ is the probability of privacy leakage.

[0018] By adopting the above technical solution, context similarity is calculated based on BERT semantic embedding, and a natural language processing model is introduced for semantic modeling; this application uses the mature and efficient TransE model to ensure the stability and computational efficiency of the embedding process, which is suitable for large-scale application of traditional Chinese medicine knowledge graphs; dynamic thresholds are set according to historical confidence distribution to improve the sensitivity and adaptability of anomaly detection; differential privacy mechanism is used for federated learning, and noise perturbation is added during the gradient update process to achieve a balance between privacy protection and model performance.

[0019] In the second aspect, the invention objective of this application is achieved by adopting the following technical solutions: The AI-based TCM syndrome differentiation-assisted diagnosis system is applied to the AI-based TCM syndrome differentiation-assisted diagnosis method described above, and the system includes: The data acquisition module is used to collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information; the pathogenesis classification mapping module is used to map the patient's symptom information to a preset TCM pathogenesis classification model to generate an initial symptom feature vector; The knowledge base module includes a knowledge base of TCM prescription rules and contraindications for drug combinations; A deep learning processing module performs data spatiotemporal alignment processing and dynamic combination inference on the initial symptom feature vector through a deep learning model to generate a personalized candidate prescription set that conforms to the TCM compatibility taboos; the dynamic combination inference adopts a solution algorithm based on the constraint satisfaction problem, and combines the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching; a prescription recommendation module is used to output a prescription medication recommendation plan including TCM compatibility, dosage and decoction method based on the candidate prescription set.

[0020] In a third aspect, the invention objective of this application is achieved by adopting the following technical solutions: A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned AI-based TCM dialectical auxiliary diagnosis method.

[0021] Fourthly, the invention objectives of this application are achieved by adopting the following technical solutions: A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the AI-based TCM dialectical auxiliary diagnosis method as described above.

[0022] In summary, this application includes at least one of the following beneficial technical effects: 1. By leveraging AI technology to enhance the effectiveness of TCM dialectical diagnosis, the system ultimately outputs accurate prescription and medication recommendations based on efficient data processing and analysis using AI technology. This helps TCM practitioners formulate final diagnosis and treatment plans, and has significant clinical application value in scenarios such as online medical platforms, remote diagnosis and treatment services, and primary healthcare institutions. 2. Context similarity is calculated based on BERT semantic embedding, and a natural language processing model is introduced for semantic modeling. This application uses the mature and efficient TransE model to ensure the stability and computational efficiency of the embedding process, which is suitable for large-scale application of traditional Chinese medicine knowledge graphs. Dynamic thresholds are set according to historical confidence distributions to improve the sensitivity and adaptability of anomaly detection. A differential privacy mechanism is used for federated learning, and noise perturbations are added during the gradient update process to achieve a balance between privacy protection and model performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of an AI-based TCM dialectical auxiliary diagnosis method in one embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application is further described in detail below with reference to the accompanying drawings.

[0025] In one embodiment, if Figure 1 As shown, this application discloses an AI-based TCM dialectical auxiliary diagnosis method, which specifically includes the following steps: S1: Collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information.

[0026] In this embodiment, a TCM syndrome differentiation assistance platform or WeChat mini-program software called "Smart TCM" has been developed. With the help of computers, "Smart TCM" has completely changed the traditional perception that TCM is difficult to learn. With diligent study, one can master thousands of years of TCM treatment experience in just a few days or even hours, facilitating the dissemination of TCM knowledge and its application.

[0027] Physiological monitoring data includes vital signs data such as blood pressure, heart rate, body temperature, and blood oxygen saturation collected through wearable devices; tongue image data is tongue images taken by high-definition cameras, including tongue color, coating, cracks and other characteristics; pulse signals are pulse waveform data collected using pressure sensors or photoplethysmography (PPG).

[0028] Specifically, patients are guided to fill in digital patient symptom information through a mobile APP to collect the main symptoms (such as bitter taste in the mouth, chest tightness, and other 360 standardized symptoms); a medical-grade tongue diagnosis instrument is used to obtain RGB images of the tongue; a wrist pulse monitoring device is used to collect 30-second pulse waveforms, and the patient's posture and movement status are simultaneously recorded.

[0029] In this embodiment, step S1 includes: S11: The patient's blood pressure, heart rate, and body temperature data are collected in real time through wearable devices, and noise filtering and missing value filling are performed to obtain preprocessed physiological monitoring data.

[0030] S12: Obtain the patient's tongue image based on the image acquisition device, use the color correction algorithm to eliminate light interference, and obtain standardized tongue image data.

[0031] In this embodiment, the image acquisition device is a multispectral tongue diagnosis instrument (including an RGB camera and a hyperspectral imaging module); the color correction algorithm is a white balance correction based on the Gray World model; and the standardized tongue image data refers to the RGB-Lab color space characteristics that comply with the "International Standard for Traditional Chinese Medicine Tongue Diagnosis" (ISO 20408:2021).

[0032] Specifically, during image acquisition, the patient's tongue was extended ≥ 2 cm, with the tip of the tongue resting against the roof of the mouth, maintaining a naturally relaxed state. For color correction, the algorithm used a grayscale plate (L95, a0, b*0) and a standard color chart (X-Rite Color Checker Classic) to calculate the average RGB values ​​(Rg, Gg, Bg) of the grayscale plate area. The color correction formulas were: Rc = R / Rg × 255; Gc = G / Gg × 255; Bc = B / Bg × 255. The highlight suppression threshold for areas with uneven illumination was: Lth = 0.7 × Lmax. The tongue was segmented using the U-Net++ model (5 layers in depth, 256 feature fusion channels). Segmentation evaluation indicators included the Dice coefficient (0.93±0.02) and IoU 0.89±0.03). Feature extraction of Lab color space parameters included: L* (lightness): 45±5 (normal tongue color); a* (red-green axis): 15±3 (reddish due to damp-heat syndrome); b* (yellow-blue axis): 18±4 (yellowish due to heat syndrome). Morphological erosion (structural element radius 3px) and distance transform algorithm were used to measure tongue coating thickness.

[0033] S13: Collect the patient's pulse signal through the pulse sensor, apply wavelet transform to remove high-frequency noise, and obtain the pulse signal.

[0034] Specifically, pulse sensors such as piezoelectric ceramic pulse wave sensors; wavelet transforms such as Daubechies wavelet (db4) time-frequency analysis; high-frequency noise including power frequency interference (e.g. 50 / 60Hz) and electromyographic artifacts (e.g. >20Hz); and the selection of suppression thresholds for high-frequency noise suppression. Where σ is the standard deviation of noise and N is the length of the pulse signal.

[0035] S2: Map the patient's symptom information to the preset TCM pathogenesis classification model to generate the initial symptom feature vector.

[0036] In this embodiment, the preset TCM pathogenesis classification model is a multi-label classification model constructed based on classic literature such as "Treatise on Febrile Diseases" and "Treatise on Warm Diseases", which includes 64 basic pathogenesis and 256 compound pathogenesis; the initial symptom feature vector includes a 128-dimensional feature vector, covering characteristics such as symptom intensity, spatiotemporal distribution, and syndrome correlation.

[0037] Specifically, based on the UMLS medical knowledge base and classic Chinese medicine literature, a knowledge graph containing 558 medicinal materials, 330 syndromes, and 900,000 compatibility relationships was constructed. The TransE algorithm was used for entity relationship embedding to generate a 128-dimensional distributed semantic representation. The pre-processed multimodal data was fused through the Transformer encoder to generate a 768-dimensional initial representation and obtain the initial symptom feature vector. The pre-trained CNN model was used to extract tongue texture features, combined with the time-frequency features of the pulse, and weighted integration was performed through the attention mechanism. The integrated features were input into the Softmax layer to output the probability distribution vectors of 64 basic pathogenesis.

[0038] S3: Call the knowledge base containing TCM prescription rules and compatibility taboos, and use the deep learning model to perform data spatiotemporal alignment processing and dynamic combination inference on the initial symptom feature vector to generate a personalized candidate prescription set that conforms to TCM compatibility taboos; the dynamic combination inference adopts a solution algorithm based on the constraint satisfaction problem, combined with the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching.

[0039] In this embodiment, data spatiotemporal alignment is to standardize data of different spatiotemporal dimensions (such as season, region, and circadian rhythm); the constraint satisfaction problem (CSP) is used to establish a mathematical model that includes constraints such as drug conflicts, dosage ranges, and incompatibility; multidimensional association rule matching refers to simultaneously considering multidimensional associations such as syndrome similarity (cosine similarity > 0.8), modern medical indicators (such as CRP levels), and physical characteristics (nine physical models).

[0040] Specifically, a knowledge base containing 300,000 historical prescription rules of traditional Chinese medicine is loaded, and each rule contains 12 attribute fields such as indications, contraindications, dosage range, and compatibility relationship; dynamic loading of real-time updated Chinese medicine pharmacopoeia data, including modern pharmacology and toxicology data of 558 medicinal materials included in the 2023 edition of the "Chinese Pharmacopoeia".

[0041] Specifically, the deep learning model is used to perform spatiotemporal data alignment and dynamic combination inference on the initial symptom feature vector, including: S301: Normalize the physiological monitoring data and construct a time series feature vector including blood pressure, heart rate, and body temperature.

[0042] In this embodiment, standardization refers to mapping physiological data of different dimensions to a unified numerical range (such as Z-score standardization); the time series feature vector is a vector representation containing trend terms, periodic terms, and statistical characteristics (mean / variance). Specifically, during data standardization, blood pressure is normalized to the interval [0, 1] using Min-Max; heart rate is logarithmically transformed to eliminate skewed distribution, and body temperature is standardized using Z-score based on human core temperature (37°C). The STL algorithm is used for time series decomposition and feature vector construction.

[0043] S302: Extract tongue coating thickness, tongue color index, and crack degree characteristic parameters of tongue image data through image segmentation algorithm to construct a tongue image feature matrix.

[0044] Specifically, the image segmentation algorithm is a U-Net++ model based on deep learning, which is used for tongue area segmentation; the tongue coating thickness refers to the vertical height of the tongue coating-covered area; the tongue color index refers to the values ​​of a* (red and green axis) and b* (yellow and blue axis) in the Lab color space; the degree of cracking refers to the area ratio of the pixel-level connected area of ​​the tongue cracks. Canny edge detection is used to extract the crack contour, and the ratio of the crack area to the total tongue area is calculated to obtain the degree of cracking.

[0045] S303: Using short-time Fourier transform to extract the time domain features and frequency domain features of the pulse signal to generate a pulse feature vector.

[0046] Specifically, short-time Fourier transform (STFT) is a time-frequency analysis tool with a window size of 512 points and an overlap rate of 50%; the time domain characteristics refer to the amplitude of the main pulse wave (H1) and the amplitude of the pre-dicrotic wave (H2); the frequency domain characteristics refer to the peak frequency of the energy density spectrum (dominant frequency: the highest frequency with an energy share of >80%).

[0047] S304: Use the dynamic time warping algorithm to align the tongue image, pulse image, and physiological monitoring features in time and space to generate a patient state vector containing multiple feature dimensions.

[0048] Specifically, dynamic time warping (DTW) is used to calculate the optimal alignment path for time series of different lengths. The DTW path calculation is subject to slope restriction constraints; spatiotemporal alignment refers to unifying the time reference of multimodal features (such as aligning tongue and pulse features to the timestamp of physiological signals).

[0049] S305: Perform weighted fusion on the multimodal features aligned by the multi-head attention mechanism to generate a comprehensive pathogenesis probability distribution matrix.

[0050] Specifically, the multi-head attention mechanism refers to the parallel calculation of multiple attention heads to capture the correlation between different features. In this embodiment, 8 attention heads are calculated in parallel; the pathogenesis probability distribution matrix includes the probability distribution of 32 basic pathogenesis (such as Qi deficiency, damp heat, and blood stasis).

[0051] Furthermore, step S3 further includes: S311: Based on the historical case database, association rule mining algorithm is used to extract frequently occurring Chinese medicine compatibility patterns.

[0052] In this embodiment, association rule mining is used to discover the correlation between variables (such as "if Astragalus is used, it is often used in combination with Angelica sinensis"); drug combination patterns with a support degree of high-frequency combination patterns ≥ a threshold (such as 5%); and a historical case database containing a structured database of patient pathogenesis labels and prescription compositions.

[0053] Specifically, the text prescription is converted into structured data (such as "Treatise on Febrile Diseases": "Guizhi Decoction" → [Guizhi, Shaoyao, Ginger, Jujube, Licorice]); a binary transaction table is constructed: patient ID → {drug 1, drug 2, ..., drug n}; and the minimum support and minimum confidence are configured, where the minimum support is set based on domain knowledge (such as 0.03); the minimum confidence is combined with clinical effectiveness verification (such as 0.6), and the association rule form between the two and the drug is set: drug A → drug B [support, confidence]; the extraction results of the association rule mining algorithm are preliminarily filtered, such as removing drug combinations containing 18 anti-drugs / 19 fears (such as Aconitum + Trichosanthes kirilowii) and retaining rules with TOP200 support and confidence > 0.7.

[0054] S312: Use machine learning models to score the effectiveness of compatibility patterns and screen out compatibility rules that meet the preset TCM pharmacopoeia standards.

[0055] Specifically, the effectiveness score refers to the quantitative indicator of comprehensive efficacy synergy and toxicity antagonism; the TCM pharmacopoeia standard refers to the incompatibility and efficacy association recorded in the "Chinese Pharmacopoeia" and "Chinese Materia Medica".

[0056] Extract drug combination characteristics, efficacy enhancement characteristics and toxicity antagonism characteristics; drug combination characteristics such as chemical composition similarity (Tanimoto coefficient); efficacy enhancement characteristics such as target enrichment analysis (DAVID database); toxicity antagonism characteristics using ADMET prediction (SwissADME platform), effectiveness threshold scoring mechanism: effectiveness score ≥ 0.6 is considered to be effective compatibility; and sort by priority: sort in descending order of score, taking the TOP30 rule.

[0057] S313: Integrate the screened compatibility rules with modern medical contraindications to construct a dynamically updated TCM knowledge graph.

[0058] In this embodiment, for example, "Intelligent Traditional Chinese Medicine" includes over 360 symptoms, more than enough to cover common illnesses. However, some symptoms cannot coexist simultaneously. For example, a white tongue coating is a sign of a cold illness, while skin erythema is a sign of a hot illness. These symptoms are incompatible, and the system excludes them. The Traditional Chinese Medicine knowledge graph is a heterogeneous graph composed of entities (drugs, pathogenesis) and relationships (compatibility taboos). The dynamic update mechanism is a scheduled update process based on PubMed literature mining.

[0059] Specifically, the entity types of the TCM knowledge graph include Chinese medicine (node ​​attributes: nature, flavor, and meridians), pathogenesis (node ​​attributes: five elements), and relationship types include compatibility enhancement (edge ​​weight: +1 to +5) and compatibility taboos (edge ​​color: red). A structured taboo table is constructed to establish the taboo rule library corresponding to "Intelligent TCM". For example, the content of the structured taboo table is as follows: Taboo Type Trigger Conditions Severity Eighteen Antidotes Aconitum + Pinellia ★★★★★ Interaction between Western medicine and pharmacology Licorice + diuretics ★★☆☆☆ Furthermore, the dynamic update mechanism of the TCM knowledge graph includes: S100: Collect patient medication feedback data and construct adverse reaction knowledge triples based on drugs, symptoms, and weight coefficients.

[0060] In this embodiment, the adverse reaction knowledge triplet is in a structured data format, such as (drug A, adverse reaction B, weight coefficient C), which represents the probability or severity of drug A causing adverse reaction B. The weight coefficient is a quantitative indicator (such as occurrence frequency or risk level) calculated based on clinical statistical data.

[0061] Specifically, medication records and adverse reaction reports were collected from the hospital's HIS system and patient follow-up platform to gather patient medication feedback data. Natural language processing was used to extract entity relationships within adverse reaction knowledge triples, filter out low-frequency triples (support <5%), and standardize the names of traditional Chinese medicines (e.g., "raw astragalus" → "Huangqi").

[0062] S200: Use the knowledge graph embedding algorithm to update the entity relationship vector and dynamically adjust the credibility of the incompatibility rules.

[0063] In this embodiment, the knowledge graph embedding algorithm is the TransE algorithm; knowledge graph embedding refers to mapping entities (drugs, pathogenesis) and relationships (compatibility) into a low-dimensional vector space.

[0064] Specifically, the TransE algorithm (suitable for ternary relationship modeling) is used: h+r≈t.

[0065] h: head entity vector (such as "Huangqi"); r: relation vector (e.g. “taboo”); t: tail entity vector (such as "licorice").

[0066] Calculate the cosine similarity of entity relationship vectors: And dynamically adjust the confidence of the taboo rule: confidence = α1×historical score + (1-α1)×embedding_similarity, where α1 is the smoothing coefficient, for example, the value is 0.7; embedding_similarity is the embedding similarity.

[0067] S300: When the confidence level of a certain incompatibility is lower than a preset confidence threshold, the expert review process is triggered for manual verification.

[0068] In this embodiment, the preset confidence threshold is a dynamic threshold; the preset confidence threshold is the minimum credibility standard set by domain experts; the manual verification process can be a multi-round expert review mechanism to ensure the scientific nature of rule revisions, and conflict resolution refers to the authoritative standards of the "Pharmacopoeia of the People's Republic of China".

[0069] S400: Integrate updated TCM clinical data based on federated learning and incrementally update the modern medicine contraindications module of the TCM knowledge graph.

[0070] In this embodiment, federated learning is a distributed machine learning framework that protects the data privacy of participating institutions such as tertiary hospitals, community clinics, and pharmaceutical R&D centers; incremental updates only synchronize model parameter changes generated by the newly added data.

[0071] S314: Generate a personalized set of candidate prescriptions that conform to TCM compatibility contraindications based on the comprehensive pathogenesis probability distribution matrix and TCM knowledge graph.

[0072] In this embodiment, the pathogenesis probability distribution matrix includes confidence vectors of 32 TCM pathogenesis (such as 0.65 for Qi deficiency and 0.32 for damp-heat); the personalized prescription set refers to the top 5 candidate prescriptions that meet the pathogenesis coverage ≥80% and the contraindication conflict number of 0.

[0073] Specifically, an optimization objective function based on maximizing the pathogenesis coverage and drug effectiveness score is constructed: Maximize the pathogenesis coverage ∑ disease coverage d ×Drug effectiveness score∑ drug_pair∈ prescription effectiveness drug-pair , the pathogenesis coverage constraint is: Ensure that the prescription covers at least 80% of the target pathogenesis set, and the incompatibility constraints are: To exclude prescriptions containing contraindicated drug pairs, where coveraged is the coverage rate of the dth pathogenesis, indicating the degree to which the prescription solves the pathogenesis; the calculation formula is coverage d =∑ drug∈prescription drug_disease_score drug,d , where drug_disease_score drug,d The contribution score of the drug to the pathogenesis d (such as the strength of the association between drug efficacy and pathogenesis). drug-pair The synergistic effectiveness score of the drug combination (drug i, drug j), i and j are different drug identifiers; the calculation formula is effectiveness drug-pair =efficacy drug-pair (1-contraindication_risk drug-pair ), where efficacy drug-pair is the expected efficacy of the drug combination (such as the success rate in historical cases), contraindication_risk drug-pair The contraindication risk of drug combinations (a probability value between 0 and 1), where the contraindication types include traditional contraindications such as the "Eighteen Antidotes" and "Nineteen Fears" (such as Aconitum + Trichosanthes), and modern contraindications. Modern contraindications include drug interactions (such as licorice + diuretics leading to electrolyte imbalance). Drug pairs are checked through knowledge graphs or TCM compatibility contraindication rule bases.

[0074] In one embodiment, dynamic combination inference uses a genetic algorithm-based constraint satisfaction problem solution, including: S321: Encode the incompatibility rules of traditional Chinese medicine and generate chromosome gene sequences.

[0075] In this embodiment, chromosome encoding refers to converting a traditional Chinese medicine prescription into a heritable digital sequence (such as a binary code or an integer code); the gene fragment of the chromosome gene sequence represents the coding unit of a single drug or a compatibility relationship.

[0076] Specifically, taboo rules are extracted from traditional Chinese medicine classics (such as "Compendium of Materia Medica") and modern guidelines, where example rules include: [('Aconite', 'Pinellia', 'Taboo'), ('Glycyrrhiza', 'Seaweed', 'Taboo')]; an encoding method is designed: for example, a unique integer ID is assigned to each traditional Chinese medicine to obtain a drug code (such as Astragalus = 0, Ginseng = 1), and a triple is used to represent the taboo relationship to obtain the taboo code. Then, a random prescription sequence (with a fixed length of N medicines) is initialized to generate a chromosome gene sequence.

[0077] S322: Evaluate the compliance of candidate prescriptions through fitness function, where compliance includes contraindication conflict rate and drug efficacy synergy.

[0078] In this embodiment, the contraindication conflict rate is the ratio of contraindication drug pairs in the prescription; and the pharmacodynamic synergy is the expected efficacy score of the drug combination.

[0079] Specifically, based on a constructed knowledge base containing Chinese medicine incompatibilities, a conflict detection algorithm is used for sliding window detection to traverse all drug pairs in the prescription, where the contraindication level weight is the penalty coefficient for different contraindications set according to the "Chinese Pharmacopoeia" and other standards. A synergy database is constructed to store the synergy effect scores of drug combinations. A graph neural network model is used to pre-train the drug synergy relationship embedding vector, and a synergy scoring algorithm is used to perform knowledge graph-based synergy scoring and GNN-based synergy scoring. The fitness function is designed based on the contraindication conflict rate and drug efficacy synergy and their respective weight coefficients. For example, the drug efficacy weight is usually set to 0.7, and the contraindication conflict rate weight is usually set to 0.3. The penalty intensity is adjusted according to the contraindication level, and rewards are given to high-confidence synergistic pairs. The specific penalty function and reward function can be defined and set.

[0080] S323: Generate new prescriptions using crossover and mutation operations, and optimize the solution space using Monte Carlo tree search.

[0081] In this embodiment, the crossover operation refers to exchanging some genes of a chromosome (such as single-point crossover and uniform crossover); the mutation operation refers to randomly changing the genes of a chromosome (such as bit flip mutation and swap mutation); and the Monte Carlo tree search (MCTS) optimizes the search path through simulation-expansion-backpropagation.

[0082] S324: Output the Top-K candidate prescription set that meets the constraint conditions.

[0083] In this embodiment, Top-K screening is to sort by fitness value and select the top K optimal solutions, where K is 30; diversity preservation refers to avoiding duplicate solutions through clustering or diversity penalty.

[0084] S4: Based on the candidate prescription set, output prescription medication recommendations including Chinese medicine compatibility, dosage and decoction method.

[0085] In this embodiment, the decoction method includes factors such as decoction utensils (casserole / ceramic), amount of water added (500 ml / dose), decoction time (15 minutes on high heat + 30 minutes on low heat), and time of taking the medicine (Chenshi / Sishi).

[0086] Specifically, prescription optimization includes: adjusting the dosage according to the patient's constitution (qi deficiency / damp-heat constitution), such as increasing the dosage of astragalus by 20% for qi deficiency and reducing the dosage of coptis chinensis by 15% for damp-heat constitution; and adjusting the medication based on seasonal factors, such as increasing the dosage of cinnamon twig by 10% in winter and increasing the dosage of honeysuckle by 15% in summer.

[0087] The generation of decoction plans includes: automatically matching the classic decoction methods recorded in "Treatise on Febrile Diseases", such as Guizhi Decoction "use seven liters of water, boil on low heat to obtain three liters, remove the residue, adjust the temperature appropriately, and take one liter"; generating personalized medical advice, including dietary taboos (such as avoiding raw, cold, and greasy foods), medication taboos (such as avoiding combined use with the Western medicine warfarin), follow-up consultation recommendations, etc.

[0088] Through the knowledge map of traditional Chinese medicine, we detect incompatibility taboos, intercept prescription combinations containing "Eighteen Antidotes" and "Nineteen Fears", and perform dosage conversion for special populations (pregnant women and those with liver and kidney dysfunction). For example, the dosage of Bupleurum is halved during pregnancy, and Alisma is contraindicated when creatinine is >133μmol / L.

[0089] In one embodiment, the dynamic update mechanism of the TCM knowledge graph further includes: S101: Perform semantic embedding of adverse reaction knowledge triples based on the BERT model and calculate the contextual similarity of drug-symptom relationships.

[0090] In this embodiment, the BERT model is a bidirectional Transformer encoder that captures semantic context through a pre-trained language model (such as Chinese BioBERT); semantic embedding refers to converting text into high-dimensional vectors while preserving semantic relationships (such as the vectors of "Huangqi + Buqi" are close); contextual similarity refers to calculating similarity based on the co-occurrence pattern of drugs and symptoms in sentences.

[0091] S102: Set a dynamic threshold τ = α × std(C) + β × mean(C), where α and β are adjustment coefficients. A warning is triggered when the confidence level C of a particular incompatibility rule is less than τ; typically, α = 1.5 and β = 2.0. Confidence level C refers to the credibility score of the incompatibility rule (range [0, 1]).

[0092] S103: Perturb the gradient update in federated learning through differential privacy mechanism, the perturbation amplitude Among them, δ is the probability of privacy leakage.

[0093] In this embodiment, differential privacy is used to add noise to prevent individual data from being reverse-engineered; gradient perturbation refers to injecting noise into the model parameter update in federated learning; and the privacy leakage probability refers to the upper limit of the probability that an attacker infers that an individual participated in training (e.g., δ = 1e-5).

[0094] It should be understood that the serial numbers of the steps in the above embodiments do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0095] In one embodiment, an AI-based TCM syndrome differentiation-assisted diagnosis system is provided. The AI-based TCM syndrome differentiation-assisted diagnosis system corresponds to the AI-based TCM syndrome differentiation-assisted diagnosis method in the above embodiment.

[0096] The AI-based TCM dialectical auxiliary diagnosis system includes a data acquisition module, a pathogenesis classification mapping module, a knowledge base module, a deep learning processing module, and a prescription recommendation module. A detailed description of each functional module is as follows: The data acquisition module is used to collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information; the pathogenesis classification mapping module is used to map the patient's symptom information to the preset TCM pathogenesis classification model to generate the initial symptom feature vector; The knowledge base module includes a knowledge base of TCM prescription rules and contraindications for drug combinations; The deep learning processing module uses a deep learning model to perform spatiotemporal data alignment and dynamic combination inference on the initial symptom feature vectors to generate a personalized set of candidate prescriptions that conform to TCM compatibility taboos. The dynamic combination inference uses a constraint satisfaction problem-solving algorithm and combines the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching. The prescription recommendation module is used to output prescription medication recommendations including Chinese medicine compatibility, dosage and decoction method based on the candidate prescription set.

[0097] For the specific limitations of the AI-based Chinese medicine dialectical auxiliary diagnosis system, please refer to the limitations of the AI-based Chinese medicine dialectical auxiliary diagnosis method above, which will not be repeated here; the various modules in the above-mentioned AI-based Chinese medicine dialectical auxiliary diagnosis system can be implemented in whole or in part through software, hardware and their combination; the above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0098] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: S1: Collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information; S2: Mapping the patient's symptom information to the preset TCM pathogenesis classification model to generate the initial symptom feature vector; S3: Calling a knowledge base containing TCM prescription rules and compatibility taboos, the deep learning model performs data spatiotemporal alignment and dynamic combination inference on the initial symptom feature vector to generate a personalized candidate prescription set that meets TCM compatibility taboos. The dynamic combination inference uses a constraint satisfaction problem-solving algorithm and combines the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching. S4: Based on the candidate prescription set, output prescription medication recommendations including Chinese medicine compatibility, dosage and decoction method.

[0099] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0100] In one embodiment, in particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, including a computer program / instruction, which implements the steps of the AI-based TCM dialectical auxiliary diagnosis method when executed by a processor. In such an embodiment, the computer program can be downloaded and installed from the network through a communication module, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the various functions defined in the present invention are performed.

[0101] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0102] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. AI-based TCM dialectical auxiliary diagnosis method, characterized by: include: Collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information; Mapping the patient's symptom information to a preset TCM pathogenesis classification model to generate an initial symptom feature vector; The knowledge base containing TCM prescription rules and compatibility taboos is called up, and the initial symptom feature vector is subjected to data spatiotemporal alignment processing and dynamic combination inference through a deep learning model to generate a personalized candidate prescription set that meets TCM compatibility taboos. The dynamic combination inference adopts a solution algorithm based on the constraint satisfaction problem, combining the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching; Based on the candidate prescription set, a prescription medication recommendation plan including Chinese medicine compatibility, dosage and decoction method is output.

2. The AI-based TCM syndrome differentiation auxiliary diagnosis method according to claim 1, characterized in that: The performing of data spatiotemporal alignment processing and dynamic combination inference on the initial symptom feature vector by a deep learning model includes: Standardizing the physiological monitoring data to construct a time series feature vector including blood pressure, heart rate, and body temperature; Extracting tongue coating thickness, tongue color index and crack degree characteristic parameters of the tongue image data through an image segmentation algorithm to construct a tongue image feature matrix; Adopting short-time Fourier transform to extract the time domain characteristics and frequency domain characteristics of the pulse signal to generate a pulse characteristic vector; The dynamic time warping algorithm is used to align tongue, pulse, and physiological monitoring features in time and space to generate a patient state vector containing multiple feature dimensions. The multimodal features aligned by the multi-head attention mechanism are weightedly fused to generate a comprehensive pathogenesis probability distribution matrix.

3. The AI-based TCM syndrome differentiation-assisted diagnosis method according to claim 2, characterized in that: The method further includes calling a knowledge base containing TCM prescription rules and compatibility taboos, performing data spatiotemporal alignment processing and dynamic combination inference on the initial symptom feature vector through a deep learning model, and further includes: Based on the historical case database, association rule mining algorithm is used to extract the frequently occurring Chinese medicine compatibility patterns; The effectiveness of the compatibility patterns is scored through machine learning models to screen out compatibility rules that meet the preset TCM pharmacopoeia standards; the screened compatibility rules are integrated with modern medical taboos to construct a dynamically updated TCM knowledge graph; Based on the comprehensive pathogenesis probability distribution matrix and the TCM knowledge graph, a personalized candidate prescription set that complies with TCM compatibility taboos is generated.

4. The AI-based TCM syndrome differentiation-assisted diagnosis method according to claim 1, characterized in that: The collecting of the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information includes: Wearable devices collect patients' blood pressure, heart rate, and body temperature data in real time, perform noise filtering and fill in missing values ​​to obtain preprocessed physiological monitoring data; The patient's tongue image is acquired based on an image acquisition device, and a color correction algorithm is used to eliminate light interference to obtain standardized tongue image data; The pulse signal of the patient is collected through the pulse sensor, and the high-frequency noise is removed by wavelet transform to obtain the pulse signal.

5. The AI-based TCM syndrome differentiation-assisted diagnosis method according to claim 4 is characterized in that: The dynamic combination inference adopts a constraint satisfaction problem solution based on a genetic algorithm, including: Encode the incompatibility rules of traditional Chinese medicine and generate chromosome gene sequences; Evaluating the compliance of candidate prescriptions by a fitness function, wherein the compliance includes contraindication conflict rate and drug efficacy synergy; Generate new prescriptions using crossover and mutation operations, and optimize the solution space using Monte Carlo tree search; Output the Top-K candidate prescription sets that meet the constraints.

6. The AI-based TCM syndrome differentiation-assisted diagnosis method according to claim 5, characterized in that: The dynamic update mechanism of the TCM knowledge graph includes: Collect patient medication feedback data and construct adverse reaction knowledge triples based on drugs, symptoms, and weight coefficients; The knowledge graph embedding algorithm is used to update the entity relationship vector and dynamically adjust the credibility of the incompatibility rules; When the confidence level of a certain incompatibility is lower than the preset confidence threshold, the expert review process is triggered for manual verification; Based on federated learning, the updated TCM clinical data is integrated and the modern medicine contraindications module of the TCM knowledge graph is incrementally updated.

7. The AI-based TCM syndrome differentiation-assisted diagnosis method according to claim 6, characterized in that: The knowledge graph embedding algorithm is the TransE algorithm; the preset confidence threshold is a dynamic threshold; and the dynamic update mechanism of the TCM knowledge graph further includes: Based on the BERT model, semantic embedding of adverse reaction knowledge triples is performed to calculate the contextual similarity of the drug-symptom relationship. A dynamic threshold τ = α × std(C) + β × mean(C) is set, where α and β are adjustment coefficients. When the confidence level C of a certain incompatibility is less than τ, an early warning is triggered. The gradient update in federated learning is perturbed through the differential privacy mechanism, and the perturbation amplitude Among them, δ is the probability of privacy leakage.

8. AI-based TCM dialectical auxiliary diagnosis system, characterized by: The system is applied to the AI-based TCM syndrome differentiation-assisted diagnosis method according to any one of claims 1 to 7, wherein the system comprises: The data acquisition module is used to collect the patient's physiological monitoring data, tongue data and pulse signals to obtain the patient's symptom information; A pathogenesis classification mapping module is used to map the patient's symptom information to a preset TCM pathogenesis classification model to generate an initial symptom feature vector; The knowledge base module includes a knowledge base of TCM prescription rules and contraindications for drug combinations; A deep learning processing module performs data spatiotemporal alignment processing and dynamic combination inference on the initial symptom feature vectors through a deep learning model to generate a personalized candidate prescription set that conforms to TCM compatibility taboos. The dynamic combination inference uses a constraint satisfaction problem-solving algorithm and combines the mapping relationship between syndromes and modern medical symptoms to perform multi-dimensional association rule matching. The prescription recommendation module is used to output a prescription medication recommendation plan including Chinese medicine compatibility, dosage and decoction method based on the candidate prescription set.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the AI-based TCM dialectical auxiliary diagnosis method as described in any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the AI-based TCM dialectical auxiliary diagnosis method as described in any one of claims 1 to 7 are implemented.

Images