Traditional Chinese medicine diagnosis and treatment suggestion system based on Al

Abstract

The invention relates to the technical field of traditional Chinese medicine diagnosis and treatment, in particular to an Al-based traditional Chinese medicine diagnosis and treatment suggestion system, which comprises a traditional Chinese medicine constitution detection module used for collecting multi-dimensional traditional Chinese medicine physical sign data of a user; the AI physique analysis engine is used for outputting quantitative scores of yang-deficiency physique and yin-deficiency physique and a mixed physique analysis result; the personalized conditioning scheme generation module is used for forming dynamically adjustable personalized health suggestions; through a multi-modal data fusion technology, face diagnosis, tongue diagnosis, pulse diagnosis and inquiry data are subjected to collaborative analysis, and a deep learning model is combined with a traditional Chinese medicine knowledge graph for comprehensive differentiation, so that error interference of a single data source is effectively reduced, the accuracy of physique classification is improved, and a diagnosis result better fits the real physique condition of a user.

Description

Technical Field

[0001] The present invention relates to the field of traditional Chinese medicine diagnosis and treatment technology, and in particular to an AI-based traditional Chinese medicine diagnosis and treatment suggestion system. Background Art

[0002] Currently, there are some TCM-assisted diagnosis systems on the market. They collect data from users on facial diagnosis, tongue diagnosis, and medical interviews, combine artificial intelligence technology to identify physical constitution, and provide conditioning suggestions. Such systems usually rely on single-modality data analysis or static knowledge bases, and lack the ability to deeply integrate and dynamically adjust multi-source information.

[0003] Existing technologies often only perform independent analysis on tongue diagnosis, pulse diagnosis or medical interview data, and lack effective cross-modal feature association methods, which may cause the constitution identification results to be affected by the errors of a single data source, reducing the robustness of the overall diagnosis. For example, when the user's tongue image deviates from the actual constitution due to short-term dietary influences, the system may not be able to combine the pulse image or medical interview data for correction, thereby providing inaccurate conditioning plans. In addition, the suggestions generated by existing systems are usually based on a fixed rule base and cannot be dynamically optimized according to changes in the user's constitution, seasonal changes or individual feedback. Long-term use may lead to diminishing conditioning effects, making it difficult to achieve accurate personalized health management.

[0004] Therefore, in response to the above problems, the present invention proposes an AI-based Chinese medicine diagnosis and treatment recommendation system, which improves the accuracy of constitution identification through a multimodal data fusion algorithm and a dynamic weight adjustment mechanism, and combines user feedback and environmental factors to achieve continuous optimization of conditioning plans, thereby improving the accuracy and adaptability of Chinese medicine health intervention. Summary of the Invention

[0005] In order to overcome the problems of insufficient multimodal data fusion and static conditioning plans in existing TCM auxiliary diagnosis systems, the present invention proposes an AI-based TCM diagnosis and treatment recommendation system.

[0006] The technical solution of the present invention is: a TCM diagnosis and treatment suggestion system based on AI, comprising:

[0007] The Traditional Chinese Medicine (TCM) constitution detection module is used to collect the user's multi-dimensional TCM physical sign data through the facial diagnosis image acquisition unit, tongue diagnosis image analysis unit, structured electronic medical consultation unit, and intelligent pulse diagnosis equipment;

[0008] The AI ​​constitution analysis engine is used to input the collected physical sign data into a deep learning model, combine it with the Traditional Chinese Medicine knowledge graph to identify constitutions, and output quantitative scores for yang deficiency and yin deficiency constitutions, as well as mixed constitution analysis results;

[0009] The personalized conditioning plan generation module is used to match the corresponding ingredient combinations from the medicine and food formula library based on the results of physical analysis, and generate physical conditioning plans such as massage and moxibustion from the external therapy database to form dynamically adjustable personalized health recommendations.

[0010] Preferably, the TCM constitution detection module includes a computer vision tongue and face diagnosis subsystem, a structured medical questionnaire unit and an intelligent pulse acquisition device. The computer vision tongue and face diagnosis subsystem is used to collect user facial and tongue images through a high-definition camera equipped with a standardized light source, and realize feature extraction and classification analysis of tongue coating color, tongue shape and facial complexion based on a convolutional neural network. The structured medical questionnaire unit is used to develop a digital interactive interface based on the TCM Ten Questions Song, collect user symptoms, living habits and medical history information through a dynamic question logic tree, and use natural language processing technology to parse text semantics. The intelligent pulse acquisition device is used to capture radial artery pulsation signals through an array pressure sensor and a three-dimensional positioning device, extract pulse position, pulse rate and pulse strength characteristic parameters in combination with a wavelet transform algorithm, and output pulse classification results.

[0011] Preferably, the AI ​​analysis engine includes multimodal data fusion technology, a dialectical reasoning module based on a knowledge graph, and a dynamic weight adjustment model. The multimodal data fusion technology is used to input tongue and facial diagnosis image features, interview text features, and pulse signal features into a unified analysis framework through a feature-level fusion algorithm. The dialectical reasoning module based on the knowledge graph is used to realize logical deduction from symptoms to constitution through a pre-constructed TCM syndrome relationship map. The dynamic weight adjustment model is used to optimize the weight distribution of each diagnostic parameter through a reinforcement learning mechanism based on the user's subsequent efficacy feedback data.

[0012] Preferably, the conditioning suggestions include a medicine-food homology scheme, an external therapy scheme and a dynamic conditioning plan module. The medicine-food homology scheme is used to generate personalized dietary suggestions including medicated meal recipes, tea substitute formulas and medicinal wine soaking schemes based on the food four properties and five flavors database and the user's physical constitution matching algorithm. The external therapy scheme is used to combine the meridian acupoint database and the user's physical sensitivity assessment to output a physical conditioning scheme including moxibustion acupoint combinations, medicinal bath formula dosages and Tai Chi move sequences. The dynamic conditioning plan module is used to adjust the food replacement suggestions and therapy intensity parameters in the treatment plan according to the time series changes of the user's health data and the solar term law prediction model.

[0013] Preferably, the medicine and food homology scheme includes an algorithm for matching the four properties of food, namely, cold, hot, warm, and cool, a logic engine for the combination of monarch, minister, assistant, and envoy, and a personalized dosage adjustment model. The algorithm for matching the four properties of food, namely, cold, hot, warm, and cool, is used to screen a combination of ingredients with suitable properties by calculating the Euclidean distance between the user's physical constitution and the nature and taste of the ingredients. The logic engine for the combination of monarch, minister, assistant, and envoy is used to generate a combination of ingredients with synergistic effects through a graph neural network based on the compatibility principles recorded in the "Compendium of Materia Medica". The personalized dosage adjustment model is used to comprehensively consider the user's weight, age, and seasonal temperature and humidity parameters to calculate the daily recommended intake threshold of each ingredient through a multiple regression equation.

[0014] Preferably, the external therapy plan includes a three-dimensional human meridian visualization positioning module and a therapy intensity adaptive algorithm. The three-dimensional human meridian visualization positioning module is used to superimpose the user's real-time body surface image with the standard meridian map through AR technology, and mark the precise body surface projection position of the recommended stimulation acupoints. The therapy intensity adaptive algorithm is used to dynamically adjust the moxibustion duration and massage intensity according to the user's pain tolerance test data and physical sensitivity score.

[0015] Preferably, the system also includes a verification mechanism for comparing the physician's diagnosis with the AI ​​results and identifying contraindications and incompatibilities.

[0016] Preferably, the verification mechanism includes a double-blind test module, an efficacy feedback learning loop and a safety warning system. The double-blind test module is used to randomly assign doctors and AI to independently diagnose the same patient during the system deployment phase, and use the Kappa coefficient to statistically compare the consistency of the physical fitness judgment results of the two. The efficacy feedback learning loop is used to update the AI ​​model weight parameters using the gradient descent method based on the symptom improvement scale data filled out by the user every week. The safety warning subsystem is used to compare the user's current plan with the contraindication database in real time, trigger a pop-up warning when a conflict is found, and lock the dangerous operation.

[0017] Preferably, the system uses blockchain evidence storage technology to generate tamper-proof hash values ​​for key nodes in the user's diagnosis and treatment process and write them into the alliance chain.

[0018] Preferably, the system also includes a privacy computing framework and a multilingual interactive interface, which is used to achieve joint modeling of "available but invisible" data from various medical institutions through federated learning technology; the multilingual interactive interface is used to integrate a multilingual reference library of traditional Chinese medicine terminology, supporting non-Chinese users to obtain localized versions of physical constitution reports through voice interaction.

[0019] Beneficial effects of the present invention:

[0020] 1. Through multimodal data fusion technology, facial diagnosis, tongue diagnosis, pulse diagnosis and medical interview data are collaboratively analyzed, and a deep learning model is combined with the traditional Chinese medicine knowledge graph for comprehensive syndrome differentiation, thereby effectively reducing the error interference of a single data source, improving the accuracy of constitution classification, and making the diagnosis results more in line with the user's actual physical condition.

[0021] 2. The present invention adopts a dynamic weight adjustment model and feedback learning mechanism, which can optimize the recommended content in real time according to the user's physical changes, seasonal changes and conditioning effect feedback, ensuring that the conditioning plan always adapts to the user's current health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Shown is a schematic diagram of the system framework of the present invention;

[0023] Figure 2 What is shown is a schematic diagram of the system workflow of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0025] See also Figure 1 The present invention provides an embodiment: a TCM diagnosis and treatment suggestion system based on AI, comprising:

[0026] The Traditional Chinese Medicine (TCM) constitution detection module is used to collect the user's multi-dimensional TCM physical sign data through the facial diagnosis image acquisition unit, tongue diagnosis image analysis unit, structured electronic medical consultation unit, and intelligent pulse diagnosis equipment;

[0027] The AI ​​constitution analysis engine is used to input the collected physical sign data into a deep learning model, combine it with the Traditional Chinese Medicine knowledge graph to identify constitutions, and output quantitative scores for nine constitutions, including Yang deficiency and Yin deficiency, as well as mixed constitution analysis results;

[0028] The personalized conditioning plan generation module is used to match the corresponding ingredient combinations from the medicine and food formula library based on the results of physical analysis, and generate physical conditioning plans such as massage and moxibustion from the external therapy database to form dynamically adjustable personalized health recommendations.

[0029] The TCM constitution detection module includes a computer vision tongue and face diagnosis subsystem, a structured medical questionnaire unit and an intelligent pulse acquisition device. The computer vision tongue and face diagnosis subsystem is used to collect user facial and tongue images through a high-definition camera equipped with a standardized light source, and realize feature extraction and classification analysis of tongue coating color, tongue shape and facial complexion based on a convolutional neural network. The structured medical questionnaire unit is used to develop a digital interactive interface based on the TCM Ten Questions Song, collect user symptoms, living habits and medical history information through a dynamic question logic tree, and use natural language processing technology to parse text semantics. The intelligent pulse acquisition device is used to capture radial artery pulsation signals through an array pressure sensor and a three-dimensional positioning device, extract pulse position, pulse rate and pulse strength characteristic parameters in combination with a wavelet transform algorithm, and output pulse classification results.

[0030] Furthermore, after the user enters the testing phase, they first complete facial frontal and lateral images and tongue upper and lower surface image acquisition in a standardized light box using a 1080P high-definition camera. The image preprocessing unit automatically corrects white balance and removes reflective interference. At the same time, the smart pulse diagnosis bracelet worn by the user collects bilateral inch, guan, and chi pulse data. The synchronously triggered electronic medical consultation system dynamically generates a questionnaire on core issues. The collected tongue and face diagnosis images are processed by the ResNet34 convolutional neural network to extract 32-dimensional feature vectors such as tongue color (RGB value), tongue coating thickness (pixel density distribution), and facial gloss. The pulse data undergoes wavelet denoising and feature extraction to output 12-dimensional parameters such as pulse position, pulse rate, and pulse strength. The medical consultation text is extracted using the BERT model for symptom keyword extraction and severity scoring. The three-modal data is standardized and spliced ​​at the feature layer and then input into the gradient boosting decision tree (GBDT) model. Combined with the pre-loaded TCM constitution knowledge graph (containing 648 judgment rules for 8 major constitution types), the final output is the probability distribution of nine constitutions and a mixed constitution score report.

[0031] The AI ​​analysis engine includes multimodal data fusion technology, a dialectical reasoning module based on knowledge graph and a dynamic weight adjustment model. The multimodal data fusion technology is used to input tongue and facial diagnosis image features, interview text features and pulse signal features into a unified analysis framework through a feature-level fusion algorithm. The dialectical reasoning module based on knowledge graph is used to realize logical deduction from symptoms to constitution through a pre-built Chinese medicine syndrome relationship map. The dynamic weight adjustment model is used to optimize the weight distribution of each diagnostic parameter through a reinforcement learning mechanism based on the user's subsequent efficacy feedback data.

[0032] Furthermore, after receiving the multimodal feature data from the TCM constitution detection module, the system first standardizes and spatially aligns the 32-dimensional feature vector of tongue diagnosis, the 12-dimensional parameters of pulse diagnosis, and the 18-dimensional symptom score of the medical interview through the feature fusion layer, and then inputs the multimodal joint analysis model based on the Transformer architecture. The model establishes the correlation weight matrix of tongue image, pulse image, and symptoms through the cross-attention mechanism, and calls the pre-built TCM knowledge graph (including 3865 dialectical rules from classics such as "Huangdi Neijing") for logical verification. In the process of deep learning reasoning, The model will sequentially perform feature importance ranking (such as determining the contribution of tongue coating thickness to the determination of phlegm-dampness constitution), constitution correlation calculation (outputting the probability distribution of nine basic constitutions through the softmax function), and mixed constitution combination optimization, and finally generate a quantitative report containing major constitution types (such as the probability of "yang deficiency constitution" 68.7%) and minor constitution types (such as the probability of "qi stagnation constitution" 32.4%), and mark key syndrome differentiation bases such as "pale, fat tongue with tooth marks + slow pulse = main symptom of yang deficiency". When receiving feedback data from users two weeks later, the model will achieve adaptive optimization by fine-tuning the network layer parameters.

[0033] The conditioning suggestions include a medicine-food homology scheme, an external therapy scheme and a dynamic conditioning plan module. The medicine-food homology scheme is used to generate personalized dietary suggestions including medicated meal recipes, tea substitute formulas and medicinal wine soaking schemes based on the food ingredients' four properties and five flavors database and the user's physical constitution matching algorithm. The external therapy scheme is used to combine the meridian acupoint database and the user's physical constitution sensitivity assessment to output a physical conditioning scheme including moxibustion acupoint combinations, medicinal bath formula dosages and Tai Chi move sequences. The dynamic conditioning plan module is used to adjust the food ingredient replacement suggestions and therapy intensity parameters in the treatment plan based on the time series changes of the user's health data and the solar term law prediction model.

[0034] The medicine and food homology scheme includes an algorithm for matching the four properties of food, namely, cold, hot, warm, and cool, a logic engine for the combination of monarch, minister, assistant, and envoy, and a personalized dosage adjustment model. The algorithm for matching the four properties of food, namely, cold, hot, warm, and cool, is used to screen food combinations with suitable properties by calculating the Euclidean distance between the user's physical constitution and the nature and taste of the food. The logic engine for the combination of monarch, minister, assistant, and envoy is used to generate a food combination scheme with synergistic effects through a graph neural network based on the compatibility principles recorded in the "Compendium of Materia Medica". The personalized dosage adjustment model is used to comprehensively consider the user's weight, age, and seasonal temperature and humidity parameters to calculate the daily recommended intake threshold of each food through a multiple regression equation.

[0035] The external therapy plan includes a three-dimensional human meridian visualization positioning module and a therapy intensity adaptive algorithm. The three-dimensional human meridian visualization positioning module is used to superimpose the user's real-time body surface image with the standard meridian map through AR technology, and mark the precise body surface projection position of the recommended stimulation acupoints. The therapy intensity adaptive algorithm is used to dynamically adjust the moxibustion duration and massage intensity according to the user's pain tolerance test data and physical sensitivity score.

[0036] The system also includes a verification mechanism for comparing physician diagnoses with AI results and identifying contraindications and incompatibilities.

[0037] The verification mechanism includes a double-blind test module, an efficacy feedback learning loop and a safety warning system. The double-blind test module is used to randomly assign doctors and AI to independently diagnose the same patient during the system deployment phase, and use the Kappa coefficient to statistically analyze the consistency of the physical fitness judgment results of the two. The efficacy feedback learning loop is used to update the AI ​​model weight parameters using the gradient descent method based on the symptom improvement scale data filled out by the user every week. The safety warning subsystem is used to compare the user's current plan with a contraindication database such as a list of prohibited acupuncture points for pregnant women in real time. When a conflict is found, a pop-up warning will be triggered and dangerous operations will be locked.

[0038] The system uses blockchain evidence storage technology to generate tamper-proof hash values ​​for key nodes in the user's diagnosis and treatment process, such as tongue images, pulse waveforms, and prescription signatures, and write them into the alliance chain.

[0039] The system also includes a privacy computing framework and a multilingual interactive interface, which is used to achieve joint modeling of "available but invisible" data from various medical institutions through federated learning technology; the multilingual interactive interface is used to integrate a multilingual reference library of traditional Chinese medicine terminology, supporting non-Chinese users to obtain localized versions of physical constitution reports through voice interaction.

[0040] Furthermore, based on the constitution diagnosis results output by the AI ​​analysis engine, the system first calls the structured medicine and food homology database (including the pharmacological properties of 1,200 kinds of food materials from classics such as "Compendium of Materia Medica") to match the four properties and five flavors, and screens out the core food combination through the graph neural network-based compatibility algorithm and generates three equivalent alternatives, such as the cinnamon-lamb-leek combination that is recommended for yang deficiency constitution, and performs secondary filtering based on the user's dietary preferences and allergy history; the external therapy plan generation module extracts the target acupoints from the meridian acupoint library according to the constitution characteristics, and uses the stimulation parameter optimization model based on reinforcement learning to calculate The system calculates the optimal moxibustion duration and massage intensity; the dynamic conditioning engine accesses local meteorological data (temperature / humidity) and user wearable device monitoring data (sleep / exercise volume), and finally integrates and generates a personalized program document containing daily medicinal diet recipes, acupoint therapy operation guides equipped with AR demonstration videos, and exercise prescriptions containing movement decomposition tutorials. The blockchain evidence storage system marks the source of Chinese medicine classics and the level of clinical evidence for each recommendation. Users can provide real-time feedback on the implementation status through the interactive interface. The system automatically generates program optimization suggestions based on weekly symptom scores, such as adjusting the amount of red dates from 15g to 20g.

[0041] See also Figure 2 , further, the workflow of the present invention is described:

[0042] The user first uses the system's high-definition camera to capture facial and tongue images under a standardized light source environment, while wearing an intelligent pulse diagnosis device to collect radial artery pulsation signals and filling out an electronic medical questionnaire. The system automatically integrates visual data, pulse waveforms, and text symptom descriptions to form an initial health record.

[0043] The computer vision module uses a pre-trained ResNet50 network to extract features such as tongue coating color distribution and facial gloss. The pulse diagnosis analysis unit decodes pulse position and pulse rate parameters through wavelet transform. The natural language processing engine identifies key symptom vocabulary from the interview text. Each module outputs standardized feature vectors for subsequent fusion analysis.

[0044] The multimodal fusion engine combines tongue features, pulse features, and symptom text features at the feature level, inputs them into an LSTM neural network with an attention mechanism, and combines them with more than 5,000 syndrome association rules in the traditional Chinese medicine knowledge graph to calculate the probability distribution of nine constitution types and generate a mixed constitution score report.

[0045] The system calls the database of medicine and food based on the physical constitution score, and uses the monarch, minister, assistant and envoy compatibility algorithm to match appropriate food combinations (such as the recommended cinnamon and mutton porridge recipe for yang deficiency constitution). At the same time, it extracts the corresponding meridian and acupoint information from the therapy library to generate moxibustion / massage plans, and automatically adjusts the dosage and frequency based on the user's BMI and seasonal factors.

[0046] Users preview the recommended acupoint locations such as Zusanli through the AR interface. The Kinect sensor monitors the completion of Ba Duan Jin movements in real time and provides correction prompts. The system dynamically updates the therapy parameter weights through reinforcement learning based on the user's score of the convenience of executing the plan and weekly symptom improvement feedback.

[0047] The present invention provides an embodiment for a standard health conditioning process:

[0048] When a 35-year-old office worker used this system, she first completed tongue and facial image acquisition, intelligent pulse diagnosis, and electronic medical questionnaire filling. AI analysis showed that she had a "Qi stagnation constitution with phlegm-dampness" (Qi stagnation score of 68%, phlegm-dampness score of 52%). The system automatically generated a conditioning plan that included a rose and tangerine peel tea recipe, Tanzhong acupoint massage instructions, and a modified Ba Duan Jin training program. The user used the phone's AR function to accurately locate the massage points. After two weeks, feedback data on improved sleep quality triggered the system to adjust the plan, adding Poria cocos and barley porridge as a breakfast alternative. Ultimately, after six weeks of conditioning, the Qi stagnation score dropped to 35%.

[0049] The present invention provides an embodiment for assisting in the management of chronic diseases:

[0050] After a 58-year-old diabetic patient connected to the system, the pulse diagnosis module detected the characteristics of a fine pulse. Combined with the main complaint of "polydipsia and polyuria" during the medical interview and the cracked tongue features, AI judged that he belonged to the "Yin deficiency and heat excess type" and generated a special plan including a recipe for bitter melon and pork ribs soup, moxibustion of the Sanyinjiao acupoint and Tai Chi cloud hand training. The system obtained the user's post-meal blood sugar data through the Bluetooth connection of the blood glucose meter and dynamically adjusted the ratio of Polygonatum sibiricum and Polygonatum odoratum in the medicinal diet. After three months, the glycated hemoglobin value dropped from 8.6% to 7.2%. During this period, the system automatically intercepted the original tea drink suggestion containing honey and replaced it with mulberry leaf and chrysanthemum drink.

[0051] The present invention provides an embodiment for special services during pregnancy and childbirth:

[0052] When a female user who is 24 weeks pregnant is found to have the physical characteristics of "spleen deficiency and dampness" through system detection, the safety warning module immediately disables the conventional plan's Zusanli acupuncture recommendation, and instead recommends a recipe for yam and red dates porridge and gentle massage of the Neiguan acupoint. In conjunction with the improved video tutorial of the Eight-Section Brocade for pregnant women, the system monitors the changes in the user's fetal heart rate during exercise through wearable devices. When frequent fetal movements are detected, the movement amplitude requirements are automatically lowered. The postpartum plan automatically switches to a Tongcao crucian carp soup lactation recipe and hot compress guidance for uterine recovery acupoints.

Claims

1. A TCM diagnosis and treatment suggestion system based on AI, characterized in that: Includes: The Traditional Chinese Medicine (TCM) constitution detection module is used to collect the user's multi-dimensional TCM physical sign data through the facial diagnosis image acquisition unit, tongue diagnosis image analysis unit, structured electronic medical consultation unit, and intelligent pulse diagnosis equipment; The AI ​​constitution analysis engine is used to input the collected physical sign data into a deep learning model, combine it with the Traditional Chinese Medicine knowledge graph to identify constitutions, and output quantitative scores for yang deficiency and yin deficiency constitutions, as well as mixed constitution analysis results; The personalized conditioning plan generation module is used to match the corresponding ingredient combinations from the medicine and food formula library based on the results of physical analysis, and generate physical conditioning plans such as massage and moxibustion from the external therapy database to form dynamically adjustable personalized health recommendations.

2. The TCM diagnosis and treatment suggestion system based on AI according to claim 1, characterized in that: The TCM constitution detection module includes a computer vision tongue and face diagnosis subsystem, a structured medical questionnaire unit and an intelligent pulse acquisition device. The computer vision tongue and face diagnosis subsystem is used to collect user facial and tongue images through a high-definition camera equipped with a standardized light source, and realize feature extraction and classification analysis of tongue coating color, tongue shape and facial complexion based on a convolutional neural network. The structured medical questionnaire unit is used to develop a digital interactive interface based on the TCM Ten Questions Song, collect user symptoms, living habits and medical history information through a dynamic question logic tree, and use natural language processing technology to parse text semantics. The intelligent pulse acquisition device is used to capture radial artery pulsation signals through an array pressure sensor and a three-dimensional positioning device, extract pulse position, pulse rate and pulse strength characteristic parameters in combination with a wavelet transform algorithm, and output pulse classification results.

3. The TCM diagnosis and treatment suggestion system based on AI according to claim 1, characterized in that: The AI ​​analysis engine includes multimodal data fusion technology, a dialectical reasoning module based on knowledge graph and a dynamic weight adjustment model. The multimodal data fusion technology is used to input tongue and facial diagnosis image features, interview text features and pulse signal features into a unified analysis framework through a feature-level fusion algorithm. The dialectical reasoning module based on knowledge graph is used to realize logical deduction from symptoms to constitution through a pre-built Chinese medicine syndrome relationship map. The dynamic weight adjustment model is used to optimize the weight distribution of each diagnostic parameter through a reinforcement learning mechanism based on the user's subsequent efficacy feedback data.

4. The TCM diagnosis and treatment suggestion system based on AI according to claim 1, characterized in that: The conditioning suggestions include a medicine-food homology scheme, an external therapy scheme and a dynamic conditioning plan module. The medicine-food homology scheme is used to generate personalized dietary suggestions including medicated meal recipes, tea substitute formulas and medicinal wine soaking schemes based on the food ingredients' four properties and five flavors database and the user's physical constitution matching algorithm. The external therapy scheme is used to combine the meridian acupoint database and the user's physical constitution sensitivity assessment to output a physical conditioning scheme including moxibustion acupoint combinations, medicinal bath formula dosages and Tai Chi move sequences. The dynamic conditioning plan module is used to adjust the food ingredient replacement suggestions and therapy intensity parameters in the treatment plan based on the time series changes of the user's health data and the solar term law prediction model.

5. The TCM diagnosis and treatment suggestion system based on AI according to claim 4, characterized in that: The medicine-food homology scheme includes an algorithm for matching the four properties of food, namely, cold, hot, warm, and cool, a logic engine for the combination of monarch, minister, assistant, and envoy, and a personalized dosage adjustment model. The algorithm for matching the four properties of food, namely, cold, hot, warm, and cool, is used to screen food combinations with suitable properties by calculating the Euclidean distance between the user's physical constitution and the nature and taste of the food. The logic engine for the combination of monarch, minister, assistant, and envoy is used to generate a food combination scheme with synergistic effects through a graph neural network based on the compatibility principles recorded in the "Compendium of Materia Medica". The personalized dosage adjustment model is used to comprehensively consider the user's weight, age, and seasonal temperature and humidity parameters to calculate the daily recommended intake threshold of each food ingredient through a multiple regression equation.

6. The AI-based TCM diagnosis and treatment suggestion system according to claim 4, characterized in that: The external therapy plan includes a three-dimensional human meridian visualization positioning module and a therapy intensity adaptive algorithm. The three-dimensional human meridian visualization positioning module is used to superimpose the user's real-time body surface image with the standard meridian map through AR technology, and mark the precise body surface projection position of the recommended stimulation acupoints. The therapy intensity adaptive algorithm is used to dynamically adjust the moxibustion duration and massage intensity according to the user's pain tolerance test data and physical sensitivity score.

7. The TCM diagnosis and treatment suggestion system based on AI according to claim 1, characterized in that: The system also includes a verification mechanism for comparing physician diagnoses with AI results and identifying contraindications and incompatibilities.

8. The AI-based TCM diagnosis and treatment suggestion system according to claim 7, characterized in that: The verification mechanism also includes a double-blind test module, an efficacy feedback learning loop and a safety warning system. The double-blind test module is used to randomly assign doctors and AI to independently diagnose the same patient during the system deployment phase, and use the Kappa coefficient to statistically analyze the consistency of the physical fitness judgment results of the two. The efficacy feedback learning loop is used to update the AI ​​model weight parameters using the gradient descent method based on the symptom improvement scale data filled out by the user every week. The safety warning subsystem is used to compare the user's current plan with the contraindication database in real time, trigger a pop-up warning when a conflict is found, and lock dangerous operations.

9. The AI-based TCM diagnosis and treatment suggestion system according to claim 1, characterized in that: The system uses blockchain evidence storage technology to generate tamper-proof hash values ​​for key nodes in the user's diagnosis and treatment process and write them into the alliance chain.

10. The AI-based TCM diagnosis and treatment suggestion system according to claim 1, characterized in that: The system also includes a privacy computing framework and a multilingual interactive interface, which is used to achieve joint modeling of "available but invisible" data from various medical institutions through federated learning technology; the multilingual interactive interface is used to integrate a multilingual reference library of traditional Chinese medicine terminology, supporting non-Chinese users to obtain localized versions of physical constitution reports through voice interaction.

Images