Personalized traditional Chinese medicine health-care intelligent recommendation system combined with acupuncture theory

Abstract

The invention relates to a personalized traditional Chinese medicine health-care intelligent recommendation system combined with an acupuncture theory. The system comprises a multi-modal data acquisition module, an AI diagnosis module and an acupuncture scheme generation module. According to the invention, the health information of the user is acquired through the multi-modal data acquisition module, the deep learning model fused with the acupuncture meridian theory is utilized to realize physique identification and disease diagnosis, and the personalized acupuncture health care scheme is generated based on the established acupuncture point, physique and disease three-dimensional association database. A reinforcement learning algorithm is combined with a user real-time feedback dynamic optimization scheme, the limitation that traditional acupuncture depends on physician experience is broken through, intelligentization, individuation and dynamics of acupuncture health care are achieved, the accuracy and operability of the health care scheme are improved, and the method is suitable for various scenes such as family health care and community medical treatment and has a wide application prospect. The method has remarkable innovativeness and practicability, the physique and the disease of the user are accurately identified based on multi-modal data, the acupuncture scheme adaptive to personal characteristics is generated, and the problem that a traditional universal scheme is insufficient in adaptability is solved.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare technology, specifically to a personalized TCM health and wellness intelligent recommendation system that incorporates acupuncture theory. Background Technology

[0002] Acupuncture, as a core component of traditional Chinese medicine, regulates the flow of qi and blood through the meridians by stimulating acupoints to achieve disease prevention and health maintenance. It has been widely applied in conditions such as insomnia, chronic fatigue syndrome, and chronic disease management. However, traditional acupuncture for health maintenance has the following limitations:

[0003] 1. Reliance on professional physician experience: The selection of acupoints, stimulation methods and parameter configurations are highly dependent on the clinical experience of physicians. Ordinary users find it difficult to operate independently, and the diagnostic results of different physicians vary, resulting in insufficient consistency and accuracy of health care plans.

[0004] 2. Low degree of personalization: Most existing acupuncture and health care programs are general and do not fully consider the differences in users' physical constitution (such as Qi deficiency, Yang deficiency, etc.), lifestyle habits and dynamic changes in health status, making it difficult to meet personalized health care needs;

[0005] 3. Lack of dynamic adjustment mechanism: Once a traditional solution is determined, it is usually not optimized in real time based on user feedback and health improvement, resulting in a gradual decrease in the adaptability of the solution;

[0006] 4. High operational threshold: Acupoint location and stimulation intensity control require professional knowledge. Ordinary users are prone to problems such as positioning deviation and improper parameters when performing the operation independently, which affects the health care effect.

[0007] Existing AI-powered TCM systems primarily focus on symptom diagnosis or herbal recommendations. While some attempt to incorporate acupuncture theory, they lack a deep understanding of the connections between constitution, acupoints, and symptoms, resulting in solutions that are not supported by TCM theory. Furthermore, they suffer from limited data collection, failing to integrate multimodal health information and lacking dynamic optimization mechanisms to adapt to changes in user health. Therefore, there is an urgent need to develop a personalized intelligent health and wellness system that integrates acupuncture theory and AI technology to address these technical challenges. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a personalized TCM health and wellness intelligent recommendation system that combines acupuncture theory. It is suitable for the intelligent generation, execution, and dynamic optimization of personalized acupuncture health and wellness plans, and solves the technical problems of traditional acupuncture health and wellness relying on experience, lacking personalization, lacking dynamic adjustment, and having high operational threshold.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a personalized TCM health and wellness intelligent recommendation system combining acupuncture theory, comprising a multimodal data acquisition module, an AI diagnosis module, an acupuncture plan generation module, a dynamic optimization module, and an interaction and execution module:

[0010] The multimodal data acquisition module is used to collect users' text-based symptom questionnaire data, image-based tongue images, facial color data, sensor-based pulse images, sleep, exercise, heart rate physiological parameters and lifestyle data. The data covers more than 20 subjective symptom indicators, more than 3 types of objective physiological parameters and more than 2 types of image data.

[0011] The AI ​​diagnostic module, based on a deep learning model that integrates acupuncture and meridian theory, adopts a three-level architecture of feature extraction, meridian constraint, and fusion diagnosis. It performs cross-dimensional feature fusion and semantic parsing on multimodal data, and completes the identification of 9 basic TCM constitutions and the correlation diagnosis of more than 50 common health care diseases.

[0012] The acupuncture plan generation module is based on a three-dimensional database of acupoints, constitution, and symptoms constructed from classic TCM literature. It matches the optimal combination of acupoints and, combined with the user's age, gender, lifestyle, and device compatibility, generates a personalized acupuncture and wellness plan that includes acupoint location, stimulation method (acupuncture, moxibustion, electroacupuncture, massage), stimulation intensity, duration, frequency, and operation contraindications.

[0013] The dynamic optimization module uses the Q-learning reinforcement learning algorithm to construct a reward function centered on "symptom improvement degree, physiological parameter optimization rate, and user satisfaction". It uses acupoint combination adjustment, stimulation parameter iteration, and execution frequency optimization as the action space, and user health status changes as the state space, forming a closed-loop iterative mechanism of diagnosis, recommendation, execution, feedback, and optimization.

[0014] The interaction and execution module supports visual display of solutions using images, text, 3D models, and AR real-scene positioning. It provides multi-channel feedback collection, including voice, star rating, and text input. It is compatible with smart acupuncture devices using Bluetooth, WiFi, and NB-IoT protocols, enabling one-click synchronous execution. It also achieves encrypted storage and authorized access to user health data through a blockchain consortium blockchain architecture.

[0015] Furthermore, the multimodal data acquisition module includes a questionnaire interaction unit, an image acquisition unit, and a sensor data interface unit;

[0016] The questionnaire interaction unit is designed based on the "Classification and Judgment Standards of Traditional Chinese Medicine Constitution" and includes 25 structured indicators such as diet, sleep, emotions, past medical history, and exercise habits. It supports dynamic adaptation of questions for different scenarios.

[0017] The image acquisition unit has a built-in standardized shooting guidance algorithm to achieve automatic focusing and light correction of tongue image (tongue tip, tongue surface, tongue root) and facial color. It completes feature preprocessing through grayscale correction, Gaussian filtering noise reduction, and region segmentation algorithms.

[0018] The sensor data interface unit is compatible with communication protocols such as Bluetooth 5.0, WiFi 6, and NB-IoT, and can connect to more than 10 mainstream health monitoring devices such as smart bracelets, pulse sensors, sleep monitors, and heart rate patches. The data is uniformly converted into JSON format and normalized.

[0019] Furthermore, the deep learning model of the AI ​​diagnostic module includes a feature extraction sub-model and a fusion diagnostic sub-model;

[0020] The feature extraction sub-model uses a ResNet50 network to extract color, texture, and morphological features from image data, a BiLSTM network to capture temporal correlation features (duration, aggravation, relief factors, and frequency of attacks) from symptom questionnaires, a fully connected network to process the numerical features of physiological parameters, and a multi-head attention mechanism to perform weighted fusion of multimodal features. The attention weights are dynamically allocated based on the correlation between features and physical condition and symptoms.

[0021] The fusion diagnostic sub-model incorporates acupuncture and meridian theory constraints, constructs a 4-dimensional correlation matrix of meridians, internal organs, constitution, and disease, performs matrix multiplication on the fusion features and the correlation matrix, and outputs the probability distribution of each constitution (accuracy to 0.01) and the correlation score of the target disease (0-100 points). The model's diagnostic accuracy is no less than 90%.

[0022] Furthermore, the three-dimensional correlation database of acupoints, constitution, and diseases includes a basic information database of acupoints, a rule database of constitution matching, a database of corresponding diseases, and a database of acupuncture parameters.

[0023] The acupoint basic information database stores the meridian affiliation, anatomical coordinates, therapeutic effects, and operational contraindications of 361 regular acupoints and 48 extra acupoints, and associates them with the corresponding meridian and organ mapping relationships.

[0024] The constitution matching rule base is based on 10,000+ clinical case data and classic TCM literature, and constructs a weighted table of acupoint preferences for 9 constitutions to clarify the matching priority of acupoints with effects such as tonifying qi, nourishing yin, and warming yang.

[0025] The disease database covers core acupoint combinations and auxiliary acupoint candidate sets for more than 50 common health care diseases such as insomnia, fatigue syndrome, and neck and shoulder pain; the acupuncture parameter database includes intensity grading (level 5 or above), duration range (5-40 minutes), and frequency range (1-7 times / week) for more than 4 stimulation methods, and marks the tolerance thresholds for different age groups and body types.

[0026] Furthermore, the reinforcement learning algorithm of the dynamic optimization module is equipped with a hierarchical reward mechanism: symptom improvement score has a weight of 0.5, physiological parameter optimization rate has a weight of 0.3, and user satisfaction score has a weight of 0.2, with reward values ​​ranging from 0 to 10 points; the action space includes three types of adjustment strategies:

[0027] Adjustments to acupoint combinations (replacing 1-2 core acupoints and adding 1-3 auxiliary acupoints), iterative stimulation parameters (intensity ±1 level, duration ±5 minutes), and optimization of execution frequency (±1 time / week).

[0028] The state space is characterized by changes in the probability distribution of physical constitution, changes in the disease correlation score, and fluctuations in physiological parameters.

[0029] When the reward value fluctuation is less than 5% in three consecutive rounds of feedback, the plan enters the stable maintenance phase. It will only be re-optimized when the change in the user's health status exceeds the preset threshold (change in physical fitness probability ≥ 0.2, change in disease correlation score ≥ 15 points).

[0030] Furthermore, the interaction and execution module includes a visualization display unit, a feedback collection unit, a smart device interface unit, and a data security unit;

[0031] The visualization unit supports AR real-scene acupoint positioning (positioning accuracy ≤1cm), 3D human body model acupoint annotation, operation step animation demonstration (adjustable from 30 seconds to 3 minutes), and provides offline download function;

[0032] The feedback collection unit is designed with 3 core indicators (symptom improvement level: ineffective, slight improvement, significant improvement, cure; ease of operation: 1-5 stars; physical tolerance: no discomfort, slight discomfort, significant discomfort), and supports 3 feedback methods: voice to text, quick star rating click, and custom text input. The feedback process takes ≤1 minute.

[0033] The intelligent device interface unit is compatible with more than 8 mainstream devices such as intelligent moxibustion devices, electroacupuncture devices, and acupoint massage devices, enabling one-click synchronization of solution parameters and real-time data transmission during execution.

[0034] The data security unit adopts a consortium blockchain architecture, where user data is stored in encrypted blocks, and the private key is kept by the user. It supports authorized access by physicians (access permissions can be set with an expiration date), which complies with the requirements of the Personal Information Protection Law and the Data Security Law.

[0035] A personalized TCM health and wellness intelligent recommendation method based on the system and incorporating acupuncture theory is characterized by the following steps:

[0036] S1: In the data acquisition phase, the multimodal data acquisition module collects 25 symptom questionnaire data, tongue image, standardized facial color image data, heart rate, pulse, sleep, exercise physiological parameters and lifestyle data from users. A preprocessing process of denoising, normalization and missing value imputation is adopted to generate a standardized health dataset.

[0037] S2: In the AI ​​diagnosis stage, the preprocessed dataset is input into a deep learning model that integrates acupuncture and meridian theory. Multimodal features are extracted by ResNet50, BiLSTM, and fully connected networks respectively. After being fused by a multi-head attention mechanism, the model is combined with the correlation matrix of meridians, internal organs, constitution, and disease to make a diagnosis and output the constitution type (including probability distribution) and disease correlation score.

[0038] S3: In the solution generation stage, based on the diagnostic results, the optimal acupoint combination is selected in the three-dimensional correlation database through a weighted scoring mechanism (physical fitness 0.4, disease correspondence 0.3, ease of operation 0.2, user tolerance 0.1). The stimulation method and parameters are determined according to the user scenario (family, community, medical), operation ability, and equipment type, and a visual acupuncture health care solution is generated.

[0039] S4: Solution Implementation and Feedback Phase. The solution is displayed through an interactive module and supports synchronous execution by smart devices. Feedback data and health status change data of users are collected 1-2 weeks after implementation.

[0040] S5: In the dynamic optimization stage, the feedback data is analyzed through the Q-learning reinforcement learning algorithm to adjust the acupoint combination, stimulation parameters or execution frequency, and then return to step S4 to form a closed-loop optimization until the scheme is stable.

[0041] Furthermore, the training process of the deep learning model in S2 incorporates an acupuncture meridian constraint loss function. This constraint loss function is constructed based on the correspondence between meridians and internal organs, and the therapeutic association between acupoints and diseases. Specifically:

[0042]

[0043] Where L cls For classification loss (cross-entropy loss), L meridian For the meridian constraint loss, α and β are weighting coefficients (α=0.7, β=0.3), L meridian The cosine similarity error between the model's output features and the correlation matrix of meridians and internal organs is used to force the model to learn feature mapping relationships that conform to traditional Chinese medicine theory.

[0044] Furthermore, the acupoint combination matching in S3 adopts a two-layer screening mechanism of core acupoints and auxiliary acupoints:

[0045] The core acupoints were selected from the disease correspondence database of the three-dimensional association database, with a matching score of ≥85 points;

[0046] Auxiliary acupoints are selected based on a constitution matching rule library, with a matching score of ≥70 points, and must meet the requirement of meridian complementarity with the core acupoints (≥1 non-same meridian).

[0047] The selection of stimulation methods follows the scenario adaptation rule of "prioritizing moxibustion or massage in home scenarios, supporting electroacupuncture or moxibustion in community scenarios, and being compatible with acupuncture or electroacupuncture in medical scenarios".

[0048] The stimulation parameters adopt adaptive adjustment logic. For those aged ≤18 years or ≥65 years, the intensity is reduced by 1 level and the duration is shortened by 5 minutes. For those with a cold constitution, the moxibustion temperature is increased by 3-5℃, and for those with a hot constitution, the moxibustion temperature is decreased by 3-5℃.

[0049] Compared with the prior art, the technical solution of this application has the following beneficial effects:

[0050] 1. This invention acquires user health information through a multimodal data acquisition module, utilizes a deep learning model integrating acupuncture and meridian theory to achieve constitution identification and disease diagnosis, generates personalized acupuncture and health care plans based on a constructed three-dimensional correlation database of acupoints, constitution, and diseases, and dynamically optimizes the plans through reinforcement learning algorithms combined with real-time user feedback. This breaks through the limitations of traditional acupuncture relying on physician experience, realizes the intelligence, personalization, and dynamism of acupuncture and health care, improves the accuracy and operability of health care plans, and is applicable to various scenarios such as family health care and community medical care, with significant innovation and practicality.

[0051] 2. This invention accurately identifies a user's constitution and symptoms based on multimodal data, generating acupuncture plans tailored to individual characteristics. This solves the problem of insufficient adaptability of traditional general plans. Through reinforcement learning algorithms combined with real-time feedback, the plan achieves closed-loop optimization, ensuring that the plan always matches the user's health status, improving the health and wellness effect. The operation process is simplified, supporting AR acupoint positioning and automated execution by smart devices. Ordinary users can complete health and wellness independently. It is suitable for various scenarios such as home, community, and workplace. The system supports adding new constitution types, symptoms, acupoints, or stimulation methods. Functional expansion can be achieved by updating the database and model parameters to adapt to more health and wellness needs. Attached Figure Description

[0052] Figure 1 This is a system framework diagram of the present invention;

[0053] Figure 2 This is a flowchart of the data acquisition process for this invention;

[0054] Figure 3 A flowchart for generating the solution of this invention;

[0055] Figure 4 This is a flowchart illustrating the dynamic optimization process of this invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1-4 This embodiment presents a personalized TCM health and wellness intelligent recommendation system that combines acupuncture theory. It includes a multimodal data acquisition module, an AI diagnosis module, an acupuncture plan generation module, a dynamic optimization module, and an interaction and execution module. The modules work together to achieve intelligent management of the entire process from data acquisition to plan optimization.

[0058] 1. Multimodal data acquisition module

[0059] This module employs a multi-dimensional data collection approach combining questionnaires, images, and sensors to comprehensively acquire user health information.

[0060] Questionnaire Interaction Unit: Based on the "Classification and Judgment Standards of Traditional Chinese Medicine Constitution", a structured questionnaire is designed, covering more than 20 indicators such as diet, sleep, emotions, and past medical history, to quickly collect users' subjective symptom information;

[0061] Image acquisition unit: Guides users to take standardized tongue images (tongue tip, tongue surface, tongue root) and facial color images through mobile terminals, and extracts objective features such as tongue color, tongue coating, and facial color through image preprocessing algorithms (grayscale correction, noise removal, region segmentation);

[0062] Sensor data interface unit: Supports communication with devices such as smart bracelets, pulse sensors, and sleep monitors to collect physiological parameters such as heart rate, pulse waveform, sleep duration, and step count in real time. The data format is uniformly JSON for easy subsequent processing.

[0063] 2. AI Diagnostic Module

[0064] This module constructs a deep learning model that integrates acupuncture and meridian theory, enabling dual functions of constitution identification and disease diagnosis.

[0065] Feature extraction sub-model: Specialized network structures are used for different types of data. The CNN network extracts the texture and color features of tongue / facial images, the BiLSTM network captures the temporal correlation features of symptom questionnaires (such as symptom duration and aggravating / relieving factors), the fully connected network processes the numerical features of physiological parameters, and the multimodal features are weighted and fused through an attention mechanism.

[0066] Fusion Diagnostic Sub-model: Introducing the constraints of acupuncture and meridian theory, constructing a correlation matrix of meridians, internal organs, constitution, and disease, performing matrix multiplication on the fusion features and the correlation matrix, and outputting the probability distribution of 9 basic constitutions (e.g., Qi deficiency constitution probability 0.85, Yang deficiency constitution probability 0.12) and the correlation score of the target disease (e.g., insomnia correlation score 0.91, fatigue syndrome correlation score 0.78).

[0067] Model training: A hybrid training method of "public dataset + clinical dataset" is adopted. The public dataset includes TCM constitution dataset and tongue image dataset. The clinical dataset consists of 10,000+ case data (including constitution type, symptoms and acupuncture plan) provided by 3 tertiary hospitals. During the training process, the acupuncture meridian constraint loss function is introduced to ensure that the diagnostic results are consistent with TCM theory.

[0068] 3. Acupuncture Plan Generation Module

[0069] This module generates personalized acupuncture and wellness plans based on diagnostic results and acupuncture theory.

[0070] Three-dimensional relational database construction: Integrating classic literature such as "The Great Compendium of Acupuncture and Moxibustion" and "Meridian Studies" with clinical data, a three-dimensional relational database of acupoints, constitutions, and diseases is constructed. It includes basic information of 361 regular acupoints (meridian affiliation, anatomical location, and main functions), acupoint matching rules for 9 constitutions (such as prioritizing acupoints for tonifying Qi such as Zusanli and Qihai for Qi deficiency), corresponding acupoint combinations for more than 50 common diseases (such as Shenmen, Neiguan, and Baihui for insomnia), and parameter ranges (intensity, duration, and frequency) for stimulation methods such as acupuncture, moxibustion, electroacupuncture, and massage.

[0071] Solution generation algorithm: A weighted scoring mechanism is used to match the optimal solution. The weighting factors include physical fitness (weight 0.4), disease correspondence (weight 0.3), ease of operation (weight 0.2), and user tolerance (weight 0.1). The specific steps are as follows:

[0072] 1) Select a set of suitable acupoints based on body constitution type;

[0073] 2) Select target acupoint combinations based on the correlation with symptoms;

[0074] 3) Determine the stimulation method based on the user scenario (home / medical) and their operational ability (e.g., moxibustion and massage are preferred for home scenarios, while electroacupuncture and acupuncture are supported for medical scenarios).

[0075] 4) Refer to the acupuncture parameter database to determine the stimulation intensity (e.g., moxibustion temperature 45-50℃), duration (15-20 minutes / session), and frequency (3-5 times / week);

[0076] Output: Generates a visual solution that includes text and image descriptions of acupoint location, animated operation steps, and detailed parameter settings, and supports printing or offline viewing.

[0077] 4. Dynamic optimization module

[0078] This module employs reinforcement learning algorithms to achieve real-time dynamic optimization of acupuncture treatment plans.

[0079] Reinforcement learning framework construction: "User health improvement" is used as the reward function, and the reward value is calculated based on the symptom improvement score, changes in physiological parameters (such as increased sleep duration and stable heart rate), and user satisfaction. The acupuncture point combination, stimulation parameters, and execution frequency of the acupuncture plan are used as the action space. The user's health status (physical probability distribution and disease correlation score) is used as the state space.

[0080] Optimization process: Every time the system collects user feedback data (1-2 weeks after execution), the reinforcement learning agent updates the action value function through the Q-learning algorithm, adjusts the acupoint combination (such as adding / replacing 1-2 auxiliary acupoints), parameter configuration (such as adjusting the moxibustion duration ±5 minutes), or execution frequency (such as increasing it by 1 time / week), to ensure that the solution always adapts to changes in the user's health status;

[0081] Convergence Mechanism: When the health improvement fluctuation is less than 5% in three consecutive rounds of feedback, the plan tends to stabilize and enters the maintenance phase. Re-optimization is only triggered when the user's health status changes significantly (such as new symptoms or changes in body type).

[0082] 5. Interaction and Execution Module

[0083] This module provides users with convenient solution interaction and execution support:

[0084] Visualization Unit: Through mobile terminal APP or mini program, the acupoint location (supports AR real-scene positioning), operation steps and precautions are displayed in the form of pictures, text, 3D models and animations, reducing the operation threshold;

[0085] Feedback Collection Unit: A simplified feedback questionnaire is designed (including three core indicators: degree of symptom improvement, ease of use, and physical tolerance), supporting rapid feedback methods such as voice input and star rating;

[0086] Intelligent device interface unit: compatible with mainstream intelligent moxibustion devices, electroacupuncture devices, and acupoint massage devices on the market, enabling one-click synchronization and automated execution of solution parameters. During device execution, data such as stimulation intensity and duration are collected in real time and fed back to the dynamic optimization module.

[0087] Data security unit: Blockchain technology is used to encrypt and store user health data, strictly adhering to privacy protection regulations, and data access is only authorized to users and professional physicians.

[0088] Example 1: Home setting - Treatment of insomnia in people with Yin deficiency constitution

[0089] User profile: 35-year-old female, working in internet operations, long-term sleep deprivation, reports difficulty falling asleep, vivid dreams and easy awakening, reddish tongue with thin coating, smart bracelet monitors sleep duration of 5.5 hours / night, resting heart rate of 78 beats / min;

[0090] Deployment devices: smartphone (iOS 16.0), smart bracelet (Huawei Watch GT4), tongue image acquisition accessory (customized supplementary lighting box), smart moxibustion device (Mijia Smart Moxibustion Box);

[0091] Execution process:

[0092] S1: Users complete 25 questionnaires through the APP, take 3 photos of their tongue and 3 photos of their facial color, and synchronize their wristband sleep / heart rate data. The APP completes data preprocessing and uploads the data to the cloud.

[0093] S2: The cloud model diagnosed it as "Yin deficiency constitution (0.82)", with a symptom correlation score of "insomnia 93 points" and related meridians "heart meridian, kidney meridian";

[0094] S3: Generated scheme: core acupoints (Shenmen, Sanyinjiao, Baihui) + auxiliary acupoints (Taixi), stimulation method "moxibustion", parameters "temperature 48℃, duration 25 minutes, frequency 3 times / week (Mon / Wed / Fri)", APP displays AR acupoint positioning and operation animation;

[0095] S4: Users locate acupoints using AR positioning, synchronize parameters to the moxibustion device, and execute automatically. Feedback is submitted after 2 weeks: "Sleep time is reduced by 30 minutes, no discomfort, easy to operate (5 stars)";

[0096] S5: Cloud computing reward value 3.55 points, moxibustion duration adjusted to 30 minutes; after another 2 weeks, user feedback "sleep duration 7.2 hours, improved dreaming", reward value 3.72 points; after another 2 weeks, feedback "sleep stable 7-8 hours", reward value 3.68 points, the plan is stable;

[0097] Results: A follow-up 6 months later showed that the user's insomnia symptoms were cured, and the physical condition test showed that the probability of "Yin deficiency constitution" was 0.65, with a 100% effective rate of health improvement.

[0098] Example 2: Community Scenario - Treatment of Neck and Shoulder Pain in Individuals with Qi Deficiency

[0099] User profile: 58-year-old male, retired teacher, who spends long hours at a desk, reports neck and shoulder pain and limited mobility. His tongue is pale with a thin white coating. His pulse sensor detects a weak pulse. His smart bracelet shows an average daily step count of 2,800.

[0100] Deployment equipment: tablet computer (Android 12.0), pulse sensor (traditional Chinese medicine intelligent pulse instrument), intelligent electroacupuncture device (configured in community health service centers);

[0101] Execution process:

[0102] S1: Community physicians assist in completing questionnaires and data collection, and synchronize pulse waveform and shoulder and neck range of motion detection data (third-party device access).

[0103] S2: Diagnosed as "Qi deficiency constitution (0.78)", with a correlation score of "neck and shoulder pain 89 points" and related meridians "large intestine meridian, Governing vessel".

[0104] S3: Generated scheme: core acupoints (Jianyu, Dazhui, Quchi) + auxiliary acupoints (Zusanli, Qihai), stimulation method "electroacupuncture + moxibustion" (adapted to community scenarios), parameters "electroacupuncture intensity level 3, duration 20 minutes, moxibustion temperature 52℃, duration 15 minutes, frequency 2 times / week";

[0105] S4: The community physician confirmed the treatment plan through the APP and performed the treatment using a smart electroacupuncture device. One month later, the patient reported that "the shoulder and neck pain was significantly reduced and the range of motion was expanded."

[0106] S5: The intensity of electroacupuncture was adjusted to level 4 via the cloud and continued for 1 month. Symptoms were basically relieved and the treatment was stable.

[0107] Results: After 2 months, the shoulder and neck pain score (VAS) dropped from 7 to 2, and the symptoms of Qi deficiency improved, with an effectiveness rate of 90%.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A personalized TCM health and wellness intelligent recommendation system combining acupuncture theory, characterized in that, It includes a multimodal data acquisition module, an AI diagnosis module, an acupuncture plan generation module, a dynamic optimization module, and an interaction and execution module. The multimodal data acquisition module is used to collect users' text-based symptom questionnaire data, image-based tongue images, facial color data, sensor-based pulse images, sleep, exercise, heart rate physiological parameters and lifestyle data. The data covers more than 20 subjective symptom indicators, more than 3 types of objective physiological parameters and more than 2 types of image data. The AI ​​diagnostic module, based on a deep learning model that integrates acupuncture and meridian theory, adopts a three-level architecture of feature extraction, meridian constraint, and fusion diagnosis. It performs cross-dimensional feature fusion and semantic parsing on multimodal data, and completes the identification of 9 basic TCM constitutions and the correlation diagnosis of more than 50 common health care diseases. The acupuncture plan generation module is based on a three-dimensional database of acupoints, constitution, and symptoms constructed from classic TCM literature. It matches the optimal combination of acupoints and, combined with the user's age, gender, lifestyle, and device compatibility, generates a personalized acupuncture and wellness plan that includes acupoint location, stimulation method (acupuncture, moxibustion, electroacupuncture, massage), stimulation intensity, duration, frequency, and operation contraindications. The dynamic optimization module uses the Q-learning reinforcement learning algorithm to construct a reward function centered on "symptom improvement degree, physiological parameter optimization rate, and user satisfaction". It uses acupoint combination adjustment, stimulation parameter iteration, and execution frequency optimization as the action space, and user health status changes as the state space, forming a closed-loop iterative mechanism of diagnosis, recommendation, execution, feedback, and optimization. The interaction and execution module supports visual display of solutions using images, text, 3D models, and AR real-scene positioning. It provides multi-channel feedback collection, including voice, star rating, and text input. It is compatible with smart acupuncture devices using Bluetooth, WiFi, and NB-IoT protocols, enabling one-click synchronous execution. It also achieves encrypted storage and authorized access to user health data through a blockchain consortium blockchain architecture.

2. The personalized TCM health and wellness intelligent recommendation system combining acupuncture theory as described in claim 1, characterized in that, The multimodal data acquisition module includes a questionnaire interaction unit, an image acquisition unit, and a sensor data interface unit. The questionnaire interaction unit is designed based on the "Classification and Judgment Standards of Traditional Chinese Medicine Constitution" and includes structured indicators such as diet, sleep, emotions, past medical history, and exercise habits. It supports dynamic and adaptive questioning for different scenarios. The image acquisition unit has a built-in standardized shooting guidance algorithm to achieve automatic focusing and light correction of tongue image (tongue tip, tongue surface, tongue root) and facial color. It completes feature preprocessing through grayscale correction, Gaussian filtering noise reduction, and region segmentation algorithms. The sensor data interface unit is compatible with Bluetooth 5.0, WiFi 6, and NB-IoT communication protocols, and can connect to mainstream health monitoring devices such as smart bracelets, pulse sensors, sleep monitors, and heart rate patches. The data is uniformly converted into JSON format and normalized.

3. The personalized TCM health and wellness intelligent recommendation system based on acupuncture theory as described in claim 1, characterized in that, The deep learning model of the AI ​​diagnostic module includes a feature extraction sub-model and a fusion diagnostic sub-model; The feature extraction sub-model uses a ResNet50 network to extract color, texture, and morphological features from image data, a BiLSTM network to capture temporal correlation features from symptom questionnaires, a fully connected network to process numerical features of physiological parameters, and a multi-head attention mechanism to perform weighted fusion of multimodal features. The attention weights are dynamically allocated based on the correlation between features and physical condition and symptoms. The fusion diagnostic sub-model incorporates acupuncture and meridian theory constraints, constructs a 4-dimensional correlation matrix of meridians, internal organs, constitution, and disease, performs matrix multiplication on the fusion features and the correlation matrix, and outputs the probability distribution of each constitution and the correlation score of the target disease. The model's diagnostic accuracy is no less than 90%.

4. The personalized TCM health and wellness intelligent recommendation system combining acupuncture theory according to claim 1, characterized in that, The three-dimensional database of acupoints, constitution, and diseases includes a basic information database of acupoints, a rule database of constitution adaptation, a database of corresponding diseases, and a database of acupuncture parameters. The acupoint basic information database stores the meridian affiliation, anatomical coordinates, therapeutic effects, and operational contraindications of 361 regular acupoints and 48 extra acupoints, and associates them with the corresponding meridian and organ mapping relationships. The constitution matching rule base is based on 10,000+ clinical case data and classic TCM literature, and constructs a weighted table of acupoint preferences for 9 constitutions to clarify the matching priority of acupoints with the effects of tonifying qi, nourishing yin, and warming yang. The disease correspondence database covers core acupoint combinations and auxiliary acupoint candidate sets for common health care diseases such as insomnia, fatigue syndrome, and neck and shoulder pain; the acupuncture parameter database contains intensity grading, duration range, and frequency interval of more than 4 stimulation methods, and marks the tolerance thresholds for different age groups and body types.

5. The personalized TCM health and wellness intelligent recommendation system combining acupuncture theory according to claim 1, characterized in that, The reinforcement learning algorithm of the dynamic optimization module is set with a hierarchical reward mechanism: symptom improvement degree weight 0.5, physiological parameter optimization rate weight 0.3, user satisfaction weight 0.2, and reward value range 0-10 points; The action space includes three types of adjustment strategies: Acupoint combination adjustment, stimulation parameter iteration, and execution frequency optimization; The state space is characterized by changes in the probability distribution of physical constitution, changes in the disease correlation score, and fluctuations in physiological parameters. When the reward value fluctuation is less than 5% in three consecutive rounds of feedback, the scheme enters a stable maintenance phase, and is only triggered to re-optimize when the change in the user's health status exceeds a preset threshold.

6. The personalized TCM health and wellness intelligent recommendation system combining acupuncture theory according to claim 1, characterized in that, The interaction and execution module includes a visualization display unit, a feedback collection unit, a smart device interface unit, and a data security unit; The visualization unit supports AR real-scene acupoint positioning, 3D human body model acupoint annotation, operation step animation demonstration, and provides offline download function; The feedback collection unit is designed with 3 core indicators and supports 3 feedback methods: voice to text, quick star rating click, and custom text input. The feedback process takes ≤1 minute. The intelligent device interface unit is compatible with mainstream devices such as intelligent moxibustion devices, electroacupuncture devices, and acupoint massage devices, enabling one-click synchronization of solution parameters and real-time data transmission during the execution process. The data security unit adopts a consortium blockchain architecture, where user data is stored in encrypted blocks, and the private key is kept by the user. It supports authorized access by physicians and complies with the requirements of the Personal Information Protection Law and the Data Security Law.

7. A personalized TCM health and wellness intelligent recommendation method based on the system described in any one of claims 1-6, incorporating acupuncture theory, characterized in that... Includes the following steps: S1: In the data acquisition phase, the multimodal data acquisition module collects 25 symptom questionnaire data, tongue image, standardized facial color image data, heart rate, pulse, sleep, exercise physiological parameters and lifestyle data from users. A preprocessing process of denoising, normalization and missing value imputation is adopted to generate a standardized health dataset. S2: In the AI ​​diagnosis stage, the preprocessed dataset is input into a deep learning model that integrates acupuncture and meridian theory. Multimodal features are extracted by ResNet50, BiLSTM, and fully connected networks respectively. After being fused by a multi-head attention mechanism, the model is combined with the correlation matrix of meridians, internal organs, constitution, and disease to make a diagnosis and output the constitution type and disease correlation score. S3: In the solution generation stage, based on the diagnostic results, the optimal acupoint combination is selected in the three-dimensional correlation database through a weighted scoring mechanism (physical fitness 0.4, disease correspondence 0.3, ease of operation 0.2, user tolerance 0.1). The stimulation method and parameters are determined according to the user scenario, operation ability, and equipment type to generate a visual acupuncture health care solution. S4: Solution Implementation and Feedback Phase. The solution is displayed through an interactive module and supports synchronous execution by smart devices. Feedback data and health status change data of users are collected 1-2 weeks after implementation. S5: In the dynamic optimization stage, the feedback data is analyzed through the Q-learning reinforcement learning algorithm to adjust the acupoint combination, stimulation parameters or execution frequency, and then return to step S4 to form a closed-loop optimization until the scheme is stable.

8. The personalized TCM health and wellness intelligent recommendation method combining acupuncture theory according to claim 7, characterized in that, The training process of the deep learning model in S2 incorporates an acupuncture meridian constraint loss function. This constraint loss function is constructed based on the correspondence between meridians and internal organs, and the association between acupoints and the indications of diseases. Specifically: Where L cls For classification loss (cross-entropy loss), L meridian For the meridian constraint loss, α and β are weighting coefficients (α=0.7, β=0.3), L meridian The cosine similarity error between the model's output features and the correlation matrix of meridians and internal organs is used to force the model to learn feature mapping relationships that conform to traditional Chinese medicine theory.

9. The personalized TCM health and wellness intelligent recommendation method combining acupuncture theory according to claim 7, characterized in that, The acupoint combination matching in S3 adopts a two-layer screening mechanism of core acupoints and auxiliary acupoints: The core acupoints were selected from the disease correspondence database of the three-dimensional association database, with a matching score of ≥85 points; Auxiliary acupoints are selected based on a constitution matching rule library, with a matching score of ≥70 points, and must meet the requirement of meridian complementarity with the core acupoints; The selection of stimulation methods follows the scenario adaptation rule of "prioritizing moxibustion or massage in home scenarios, supporting electroacupuncture or moxibustion in community scenarios, and being compatible with acupuncture or electroacupuncture in medical scenarios". The stimulation parameters adopt adaptive adjustment logic. For those aged ≤18 years or ≥65 years, the intensity is reduced by 1 level and the duration is shortened by 5 minutes. For those with a cold constitution, the moxibustion temperature is increased by 3-5℃, and for those with a hot constitution, the moxibustion temperature is decreased by 3-5℃.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the personalized TCM health and wellness intelligent recommendation method combining acupuncture theory as described in any one of claims 7-9.

Images