An intelligent recommendation system for traditional Chinese medicine meridian therapy solutions for chronic pain
By combining multimodal data and attention networks with an intelligent recommendation system, a personalized Traditional Chinese Medicine meridian therapy plan is generated, solving the problem of traditional methods relying on experience and wearing digital devices, and achieving efficient and accurate chronic pain management.
Patent Information
- Application Number
- CN202510024121.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-07
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2045-01-07
AI Technical Summary
Existing chronic pain management technologies rely on physician experience, are inefficient and lack personalized analysis. Digital devices need to be worn regularly, affecting daily life, and data collection lacks depth and personalization.
An intelligent recommendation system for Traditional Chinese Medicine meridian therapy plans for chronic pain is designed, including multimodal data collection, knowledge graph construction, intelligent analysis and therapy plan recommendation modules. Multimodal attention network and optimization calibration network are used to generate personalized treatment plans, reducing dependence on physicians and equipment wearing.
It improves consultation efficiency and diagnostic accuracy, provides personalized treatment recommendations, reduces the burden on doctors, and allows patients to quickly obtain treatment recommendations. It is simple to operate and easy to promote.
Smart Images
Figure CN119851866B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of deep learning and TCM-assisted diagnosis and treatment technology, and more specifically, to an intelligent recommendation system for TCM meridian therapy plans for chronic pain. Background Art
[0002] According to the "China Pain Medicine Development Report (2020)," the number of chronic pain patients in my country has reached 300 million and continues to grow at a rate of 10 to 20 million per year. Traditional Chinese medicine (TCM) therapies such as acupoint moxibustion or electrical stimulation can effectively alleviate chronic pain. Accurate recommendations for highly precise and personalized TCM meridian therapy plans will help improve the effectiveness of TCM meridian therapy. In the field of chronic pain management, traditional TCM consultation methods, represented by acupuncture and massage, rely heavily on the physician's experience and skills. For example, Bai's skin acupuncture therapy, a pain treatment method based on TCM theory, uses a comprehensive symptom-based interview and myofascial trigger point localization to determine the patient's pain type and severity. This TCM treatment approach combines meticulous interviews and palpation to quickly locate pain and provide personalized treatment plans based on Bai's acupuncture experience. However, existing myofascial trigger point localization methods rely heavily on interviews and palpation, requiring not only high physician experience and skills but also significant time consumption, making them unable to meet the needs of the growing number of pain patients. In addition, differences in patients' pain sensitivity or subjective description errors during palpation may affect diagnostic accuracy.
[0003] With the advancement of technology, a variety of rehabilitation and physical therapy products using digital health monitoring devices have emerged on the market. For example, some products integrate wearable devices that are installed in fixed locations on the body. They can monitor patients' physiological parameters in real time and provide pain management recommendations through data analysis. The function of these technologies is to assist physicians in diagnosis and treatment through objective physiological data.
[0004] Existing chronic pain management technologies, both traditional Chinese medicine and modern scientific approaches, have limitations. The former relies heavily on the physician's personal experience, techniques, and the patient's subjective statements, and is inefficient in tailoring consultation plans to individual differences. While the latter provides objective data, they require regular wear in a fixed location, disrupting daily life. Furthermore, the data collection and analysis process often lacks personalized and in-depth symptom analysis. Therefore, there is an urgent need to develop an intelligent consultation and recommendation system that combines patient information with physician knowledge to improve both efficiency and accuracy. Summary of the Invention
[0005] In order to overcome the defects of the traditional methods in the above-mentioned prior art, such as excessive reliance on manual experience, strong subjectivity, low consultation efficiency, the need to wear digital devices regularly, and the lack of personalized and in-depth symptom analysis, the present invention provides an intelligent recommendation system for Chinese medicine meridian therapy plans for chronic pain. While realizing personalized consultation, it improves consultation efficiency and reduces the burden on doctors. There is no need to wear fixed equipment to collect data, and the operation is simple and easy to promote. Non-professionals can also conduct independent pain consultation and rehabilitation management.
[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:
[0007] An intelligent recommendation system for TCM meridian therapy plans for chronic pain, including: a multimodal data acquisition module, a knowledge graph construction module, an intelligent analysis module, and a therapy plan recommendation module;
[0008] The multimodal data acquisition module is connected to the intelligent analysis module and the physical therapy plan recommendation module in sequence; the knowledge graph construction module is connected to the physical therapy plan recommendation module;
[0009] The multimodal data acquisition module is used to collect and pre-process multimodal data of chronic pain patients, wherein the multimodal data includes at least: medical history information, pain location, pain type and personalized action examination feedback data;
[0010] The knowledge graph construction module is used to construct acupoint search knowledge graph and human body link relationship graph based on preset TCM meridian therapy experience data;
[0011] The intelligent analysis module is pre-set with a multimodal attention network, which is used to perform feature extraction, feature fusion and preliminary acupoint prediction on the pre-processed multimodal data to generate a preliminary set of associated acupoints;
[0012] The therapy plan recommendation module is used to obtain a general therapy plan based on the acupuncture point search knowledge graph, and input the general therapy plan, the preliminary associated acupuncture point set and the human body link relationship diagram into a preset optimization calibration network, and finally generate a personalized Chinese medicine meridian therapy plan to complete intelligent recommendation.
[0013] Preferably, the multimodal data is obtained based on the patient's complaint, and the medical history information includes at least: pain area complaint information, accompanying pain area complaint information, pain intensity complaint information, pain duration complaint information, pain history complaint information, treatment records, and medication use;
[0014] The pain site includes at least one of the head, neck, shoulder blade, chest, back, waist, pelvis and hip, leg, knee, ankle, foot, elbow, wrist or finger;
[0015] The pain types include: dull pain or sharp pain, wherein dull pain includes at least soreness, distending pain and dull pain, and sharp pain includes at least tearing pain, cutting pain, stabbing pain, burning pain and colic;
[0016] Based on the pain area selected by the patient, the patient completes the movement examination in the horizontal, coronal and sagittal planes according to the preset movement prompts, and performs a VAS score on each movement. The VAS score is used as the patient's personalized movement examination feedback data.
[0017] Preferably, the preprocessing includes: information encoding of the multimodal data, wherein the information encoding includes mapping and standardizing the multimodal data according to a predefined data model;
[0018] The multimodal data after information encoding is preprocessed by removing duplicate data, filling missing values, correcting format errors and processing outliers.
[0019] Preferably, in the knowledge graph construction module, preset TCM meridian therapy experience data is imported into the graph database to form an acupoint-finding knowledge graph;
[0020] In the knowledge graph construction module, based on the preset traditional Chinese medicine meridian therapy experience data and according to the strength characteristics of the human body link association, an adjacency matrix diagram containing at least 90 mutual strength relationships between diagnosis and treatment points is constructed to obtain a human body link relationship diagram; in the human body link relationship diagram, the nodes of the diagram represent different acupuncture points, and the edges represent the direct association relationship between acupuncture points.
[0021] Preferably, the preset TCM meridian therapy experience data is specifically medical record data related to TCM meridian therapy obtained from multiple sources.
[0022] Preferably, in the intelligent analysis module, the structure of the multimodal attention network includes: an embedding layer, a multi-head self-attention layer, a mapping layer, a mutual attention layer and a first multi-layer perceptron layer connected in sequence;
[0023] The embedding layer is used to extract key features from the pre-processed multimodal data, map the discrete variables in the pre-processed multimodal data into a continuous vector space, and obtain a multimodal embedding vector;
[0024] The multi-head self-attention layer is used to perform adaptive feature weighting on the multimodal embedding vector;
[0025] The mapping layer is used to project the multimodal embedding vector after the multi-head self-attention layer into a higher-dimensional vector space to obtain a multimodal high-dimensional vector;
[0026] The mutual attention layer performs feature fusion on the multimodal high-dimensional vector based on the mutual attention mechanism, captures the mutual dependence between different modalities by calculating the similarity weights between each modality, and generates fusion features after weighting;
[0027] The first multi-layer perceptron layer is used to make preliminary probability predictions of acupoints based on the fusion features, and to group acupoints with the top several predicted probabilities into a preliminary associated acupoint set for output.
[0028] Preferably, in the therapy plan recommendation module, the optimization calibration network includes: a graph attention layer and a second multi-layer perceptron layer connected in sequence;
[0029] The universal physiotherapy plan, the preliminary associated acupoint set, and the human body link relationship graph are input into the graph attention layer. The graph attention layer is used to use the human body link relationship graph to perform relationship modeling and semantic enhancement on the universal physiotherapy plan and the preliminary associated acupoint set, and calculate the deep nonlinear features of the physiotherapy plan containing the human body link relationship;
[0030] The second multi-layer perceptron layer is used to optimize and calibrate deep nonlinear features, capture the implicit combination patterns of acupuncture points, output the final acupuncture point prediction results, and save them as personalized Chinese medicine meridian therapy plans to complete intelligent recommendations.
[0031] Preferably, the system further comprises: a feedback update module; the feedback update module is connected to the knowledge graph construction module, the intelligent analysis module and the physical therapy plan recommendation module respectively;
[0032] The feedback update module is used to update the cave-finding knowledge graph and fine-tune the multimodal attention network and the optimization calibration network.
[0033] Preferably, chronic pain patients receive treatment according to a personalized TCM meridian therapy plan, and feedback efficacy data is sent to the feedback update module;
[0034] The preset loss function is calculated based on the therapeutic effect data, the acupuncture knowledge graph is updated according to the preset loss function, and the multimodal attention network and the optimization calibration network are fine-tuned.
[0035] Preferably, the preset loss function is specifically:
[0036] Loss=minΔRI(p)
[0037] Among them, Loss is the preset loss function value; ΔRI(p) is the recovery index obtained based on the efficacy data.
[0038] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:
[0039] The present invention provides an intelligent recommendation system for traditional Chinese medicine meridian therapy plans for chronic pain, comprising: a multimodal data acquisition module, a knowledge graph construction module, an intelligent analysis module, and a therapy plan recommendation module; the multimodal data acquisition module is used to collect multimodal data of chronic pain patients and perform preprocessing; the knowledge graph construction module is used to construct an acupuncture point search knowledge graph and a human body link relationship diagram based on preset traditional Chinese medicine meridian therapy experience data; a multimodal attention network is preset in the intelligent analysis module, and the multimodal attention network is used to perform feature extraction, feature fusion, and preliminary prediction of acupuncture points on the preprocessed multimodal data to generate a set of preliminary associated acupuncture points; the therapy plan recommendation module is preset with an optimization calibration network, which is used to use the general therapy plan given by the acupuncture point search knowledge graph, the preliminary associated acupuncture point set calculated by the model, and the human body link relationship diagram to generate a personalized traditional Chinese medicine meridian therapy plan and complete intelligent recommendation;
[0040] By combining multimodal data and attention-based neural networks, the present invention can more accurately identify the location and type of pain and provide personalized Chinese medicine meridian therapy plans. The recommendation accuracy of this system is comparable to that of professional doctors' diagnoses, making treatment more targeted. At the same time, this system reduces the communication time between doctors and patients in traditional consultations through automated consultation and movement inspection processes, reduces the burden on doctors, and patients can also get treatment suggestions more quickly, reducing waiting time, thereby effectively improving consultation efficiency. In addition, based on the constructed knowledge graph, this system can simulate the doctor's consultation experience and provide professional treatment suggestions. There is no need to wear fixed equipment to collect data. It is simple to operate and easy to promote. This means that in some cases, ordinary people can also use the system for self-treatment or auxiliary treatment, reducing dependence on professional doctors. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 This is a structural diagram of an intelligent recommendation system for Traditional Chinese Medicine meridian therapy solutions for chronic pain provided in Example 1.
[0042] Figure 2 This is a structural diagram of an intelligent recommendation system for Traditional Chinese Medicine meridian therapy solutions for chronic pain provided in Example 2.
[0043] Figure 3 This is a workflow diagram of an intelligent recommendation system for Traditional Chinese Medicine meridian therapy solutions for chronic pain provided in Example 2.
[0044] Figure 4 This is a schematic diagram of the system model pre-training provided in Example 2.
[0045] Figure 5 This is a schematic diagram of the human body movement inspection provided in Example 2.
[0046] Figure 6 Schematic diagram of the multi-head self-attention layer provided in Example 2.
[0047] Figure 7 This is a schematic diagram of the alignment of the patient complaint data provided in Example 2.
[0048] Figure 8 This is a schematic diagram of the acupoint-finding knowledge graph provided in Example 2.
[0049] Figure 9 This is a schematic diagram of the human body link relationship provided in Example 2.
[0050] Figure 10 This is a schematic diagram of the optimized calibration network structure and reasoning process provided in Example 2.
[0051] Figure 11 Schematic diagram of the feedback-based knowledge graph update and model parameter fine-tuning provided in Example 2.
[0052] Figure 12 This is a comparison chart of the gender and age group statistics (number of people) of the experimental group and the control group provided in Example 3.
[0053] Figure 13 This is a comparison chart of the pain location statistics (number of people) of the experimental group and the control group patients provided in Example 3.
[0054] Figure 14 This is a schematic diagram of the intelligent Chinese medicine meridian therapy instrument provided in Example 3. DETAILED DESCRIPTION
[0055] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;
[0056] In order to better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product size;
[0057] It is understandable to those skilled in the art that some well-known structures and descriptions thereof may be omitted in the drawings.
[0058] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.
[0059] Example 1
[0060] like Figure 1 As shown, this embodiment provides an intelligent recommendation system for traditional Chinese medicine meridian therapy plans for chronic pain, including: a multimodal data acquisition module, a knowledge graph construction module, an intelligent analysis module and a therapy plan recommendation module;
[0061] The multimodal data acquisition module is connected to the intelligent analysis module and the physical therapy plan recommendation module in sequence; the knowledge graph construction module is connected to the physical therapy plan recommendation module;
[0062] The multimodal data acquisition module is used to collect and pre-process multimodal data of chronic pain patients, wherein the multimodal data includes at least: medical history information, pain location, pain type and personalized action examination feedback data;
[0063] The knowledge graph construction module is used to construct acupoint search knowledge graph and human body link relationship graph based on preset TCM meridian therapy experience data;
[0064] The intelligent analysis module is pre-set with a multimodal attention network, which is used to perform feature extraction, feature fusion and preliminary acupoint prediction on the pre-processed multimodal data to generate a preliminary set of associated acupoints;
[0065] The therapy plan recommendation module is used to obtain a general therapy plan based on the acupuncture point search knowledge graph, and input the general therapy plan, the preliminary associated acupuncture point set and the human body link relationship diagram into a preset optimization calibration network, and finally generate a personalized Chinese medicine meridian therapy plan to complete intelligent recommendation.
[0066] During implementation, the system constructs acupuncture point knowledge graphs based on traditional Chinese medicine meridian therapy experience data. It also converts human body link relationships into mathematical language understandable to the network, forming human body link reasoning rules. The system then combines the patient's motion examination, medical history information, and the associated acupuncture point set calculated by the model based on the aforementioned knowledge graphs and human body link reasoning rules to ultimately form a personalized treatment plan.
[0067] This system is centered around the fusion analysis of multimodal data. Through multimodal feature extraction and fusion, it accurately analyzes the patient's pain characteristics and optimizes and calibrates personalized consultation plans by combining knowledge graphs and human body association link diagrams. The detailed steps are as follows:
[0068] The multimodal data collection module is used to collect multimodal data such as the patient's chief complaint, movement examination, pain level and pain coronal plane;
[0069] 1) Chief complaint information: Patients enter their chief complaint into the electronic medical record, including the main pain area, accompanying pain area, pain intensity, pain duration, and previous pain history, to initially establish the patient's electronic medical record;
[0070] 2) Movement test: pain and functional impairment are assessed by having the patient perform specific movements and pain is assessed using a visual analogue scale (VAS) to quantify pain intensity;
[0071] VAS uses a straight line (usually 10 cm or 100 mm in length) to indicate the degree of pain, with one end of the line representing "no pain" and the other end representing "the most severe pain"; patients mark the corresponding position on the line according to the degree of pain they feel; medical staff assess the degree of pain based on the position of the patient's mark. The result of the VAS score can be a continuous value, so it can reflect the changes in pain in more detail; VAS score usually divides pain into 10 levels, with 0 indicating no pain, 1-3 points for mild pain, 4-6 points for moderate pain, and 7-10 points for severe pain; this score is used to quantify the changes in pain before and after treatment, intuitively evaluate the treatment effect of the intelligent Chinese medicine physiotherapy device, and verify the accuracy of the system-generated plan;
[0072] Based on the clinical experience and knowledge of professional physicians, the above information is converted into model input, which is divided into the following steps:
[0073] Preprocessing: Digitally encode multimodal data, including the patient's chief complaint, motor examination results, pain intensity, and pain distribution. For example, the pain location can be encoded as a number, and the pain intensity can be quantified using a VAS score. After encoding, the data is input into the multimodal attention network in the intelligent analysis module.
[0074] Feature extraction: Utilizes a multimodal attention network to extract key features from the encoded information; for example, extracting abnormal values of joint range of motion from movement tests, and extracting the main area of pain from pain distribution;
[0075] Feature weighting: Utilizes a multi-head self-attention mechanism to adaptively increase the network's focus on key information, such as joint inspection during joint pain.
[0076] Feature fusion: Based on the mutual attention mechanism, different pain syndrome information encoding vectors are fused;
[0077] Preliminary prediction: The fused features are projected and dimensionally reduced through a multi-layer perceptron to output a unique set of preliminary associated acupuncture points that match the patient's current condition;
[0078] The knowledge graph construction module uses the experience of professional physicians to build a knowledge graph, linking the location of pain, the nature of pain, and the results of movement tests with possible treatment plans. For example, if a patient complains of low back pain and the pain is distributed on the coronal plane, the knowledge graph can search for a universal treatment plan that is suitable for patients with low back pain and pain distribution on the coronal plane. At the same time, this module also summarizes Bai's experience to derive a human body link association diagram.
[0079] The unique set of preliminary associated acupoints that match the current patient's condition (the acupoints associated with the current treatment of this patient's condition) output by the intelligent analysis module is combined with the general treatment plan (without personalized differences) searched from the acupoint search knowledge graph and the human body link association relationship graph to generate a personalized treatment plan. In this embodiment, a graph attention network is used to optimize and correct the acupoint features and ultimately output a personalized treatment plan to ensure that the treatment plan meets the patient's specific physical condition.
[0080] This system provides a more accurate and personalized treatment plan. While achieving personalized consultation, it improves consultation efficiency and reduces the burden on doctors. There is no need to wear fixed equipment to collect data during the movement inspection stage. The whole process is simple to operate and easy to promote. Non-professionals can also conduct independent pain consultation and rehabilitation management.
[0081] Example 2
[0082] like Figure 2 As shown, this embodiment provides an intelligent recommendation system for traditional Chinese medicine meridian therapy plans for chronic pain, including: a multimodal data acquisition module, a knowledge graph construction module, an intelligent analysis module, a therapy plan recommendation module and a feedback update module;
[0083] The multimodal data acquisition module is connected to the intelligent analysis module and the physical therapy plan recommendation module in sequence; the knowledge graph construction module is connected to the physical therapy plan recommendation module;
[0084] The multimodal data acquisition module is used to collect and pre-process multimodal data of chronic pain patients, wherein the multimodal data includes at least: medical history information, pain location, pain type and personalized action examination feedback data;
[0085] The knowledge graph construction module is used to construct acupoint search knowledge graph and human body link relationship graph based on preset TCM meridian therapy experience data;
[0086] The intelligent analysis module is pre-set with a multimodal attention network, which is used to perform feature extraction, feature fusion and preliminary acupoint prediction on the pre-processed multimodal data to generate a preliminary set of associated acupoints;
[0087] The therapy plan recommendation module is used to obtain a general therapy plan based on the acupuncture point search knowledge map, and input the general therapy plan, the preliminary associated acupuncture point set, and the human body link relationship diagram into a preset optimization calibration network to ultimately generate a personalized Chinese medicine meridian therapy plan and complete intelligent recommendation;
[0088] The feedback update module is used to update the cave-finding knowledge graph and fine-tune the multimodal attention network and the optimization calibration network;
[0089] In this embodiment, chronic pain patients are treated according to a personalized TCM meridian therapy plan and feedback efficacy data is sent to the feedback update module;
[0090] Calculate the preset loss function based on the therapeutic effect data, update the acupuncture knowledge graph based on the preset loss function, and fine-tune the multimodal attention network and optimization calibration network;
[0091] The preset loss function is specifically:
[0092] Loss=minΔRI(p)
[0093] Among them, Loss is the preset loss function value; ΔRI(p) is the recovery index obtained based on the efficacy data;
[0094] The multimodal data is obtained based on the patient's complaint, wherein the medical history information includes at least: information on the main complaint of pain area, information on the main complaint of accompanying pain area, information on the main complaint of pain intensity, information on the main complaint of pain duration, information on the main complaint of previous pain history, treatment records, and medication use;
[0095] The pain site includes at least one of the head, neck, shoulder blade, chest, back, waist, pelvis and hip, leg, knee, ankle, foot, elbow, wrist or finger;
[0096] The pain types include: dull pain or sharp pain, wherein dull pain includes at least soreness, distending pain and dull pain, and sharp pain includes at least tearing pain, cutting pain, stabbing pain, burning pain and colic;
[0097] Based on the pain area selected by the patient, the patient completes the movement test in the horizontal, coronal and sagittal planes according to the preset movement prompts, and performs a VAS score for each movement. The VAS score is used as the patient's personalized movement test feedback data;
[0098] The preprocessing includes: performing information encoding on the multimodal data, wherein the information encoding includes mapping and standardizing the multimodal data according to a predefined data model;
[0099] The multimodal data after information encoding is preprocessed by removing duplicate data, filling missing values, correcting format errors and processing outliers;
[0100] In the knowledge graph construction module, the preset TCM meridian therapy experience data is imported into the graph database to form an acupoint-finding knowledge graph;
[0101] In the knowledge graph construction module, based on the preset TCM meridian therapy experience data and the strength characteristics of the human body link association, an adjacency matrix diagram containing the mutual strength relationships between at least 90 diagnosis and treatment points is constructed to obtain a human body link relationship diagram; in the human body link relationship diagram, the nodes of the diagram represent different acupuncture points, and the edges represent the direct associations between acupuncture points;
[0102] The preset TCM meridian therapy experience data is specifically medical record data related to TCM meridian therapy obtained from multiple sources; in this embodiment, the medical record data related to TCM meridian therapy is obtained from two sources: the hospital's online system and the doctor's offline interviews; the TCM meridian therapy experience data in this embodiment is specifically related to medical record data of Bai's skin acupuncture therapy;
[0103] In the intelligent analysis module, the structure of the multimodal attention network includes: an embedding layer, a multi-head self-attention layer, a mapping layer, a mutual attention layer, and a first multi-layer perceptron layer connected in sequence;
[0104] The embedding layer is used to extract key features from the pre-processed multimodal data, map the discrete variables in the pre-processed multimodal data into a continuous vector space, and obtain a multimodal embedding vector;
[0105] The multi-head self-attention layer is used to perform adaptive feature weighting on the multimodal embedding vector;
[0106] The mapping layer is used to project the multimodal embedding vector after the multi-head self-attention layer into a higher-dimensional vector space to obtain a multimodal high-dimensional vector;
[0107] The mutual attention layer performs feature fusion on the multimodal high-dimensional vector based on the mutual attention mechanism, captures the mutual dependence between different modalities by calculating the similarity weights between each modality, and generates fusion features after weighting;
[0108] The first multi-layer perceptron layer is used to make preliminary probability predictions of acupoints based on the fusion features, and to group acupoints with the top predicted probabilities into a preliminary associated acupoint set for output;
[0109] In the therapy plan recommendation module, the optimization calibration network includes: a graph attention layer and a second multi-layer perceptron layer connected in sequence;
[0110] The universal physiotherapy plan, the preliminary associated acupoint set, and the human body link relationship graph are input into the graph attention layer. The graph attention layer is used to use the human body link relationship graph to perform relationship modeling and semantic enhancement on the universal physiotherapy plan and the preliminary associated acupoint set, and calculate the deep nonlinear features of the physiotherapy plan containing the human body link relationship;
[0111] The second multi-layer perceptron layer is used to optimize and calibrate deep nonlinear features, capture the implicit combination patterns of acupuncture points, output the final acupuncture point prediction results, and save them as personalized Chinese medicine meridian therapy plans to complete intelligent recommendations.
[0112] In the specific implementation process, Figure 3 As shown, this system constructs an acupuncture point knowledge graph based on the experience data of traditional Chinese medicine meridian therapy (this embodiment uses the experience of Bai's therapy, which achieves the effect of relieving pain and regulating meridian qi and blood by applying weak electrical signals to specific acupuncture points on the skin surface. It is suitable for a variety of chronic pain problems, such as neuralgia, stomachache, biliary colic, etc.), and at the same time converts the experience data into a mathematical language that can be understood by the network to form a human body link reasoning rule based on Bai's therapy experience. The system realizes autonomous learning of the reasoning relationship between acupuncture point treatment points based on the aforementioned knowledge graph, and finally forms a personalized treatment plan by combining the patient's movement examination and the main complaint medical record information;
[0113] Before making a formal recommendation, Figure 4 As shown, the neural network model in this system needs to be pre-trained. The process is as follows: based on a large number of treatment cases of doctors, the model in the system is iteratively trained. The difference between the manually determined physical therapy plan and the personalized TCM meridian physical therapy plan generated by the system is calculated according to the similarity loss function. The loss function uses the cross entropy loss. During the training phase, the system ensures that after learning from a large number of cases, the recommendation system can achieve physical therapy plan recommendations that are basically consistent with those of doctors.
[0114] The system is centered around the fusion analysis of multimodal data. Through multimodal feature extraction and data fusion, it accurately analyzes the patient's pain characteristics and optimizes and calibrates personalized consultation plans by combining knowledge graphs and graph neural networks. The detailed steps are as follows:
[0115] First, a multimodal data collection module is used to collect multimodal data from patients with chronic pain, including medical history information, pain location, pain type, and personalized movement test feedback data;
[0116] In this embodiment, multimodal data is constructed based on the patient's complaints. The medical history information includes at least: information on the main complaint area of pain, information on the main complaint area of accompanying pain, information on the main complaint area of pain intensity, information on the main complaint area of pain, information on the main complaint area of pain history, treatment records, and medication usage. The recommendation system can use this information to comprehensively understand the patient's health status and the individual differences in the benefits of the consultation plan, thereby formulating a personalized treatment plan.
[0117] Pain location refers to the distribution and manifestation of pain on different planes of the body. This embodiment divides the human body into 14 major segments, including the head, neck, shoulder blades, chest, back, waist, pelvis and hips, legs, knees, ankles, feet, elbows, wrists, and fingers. Understanding the pain location helps the recommendation system locate the source of the pain and recommend appropriate treatment options based on the source.
[0118] Pain types include dull pain and sharp pain. Dull pain includes at least aching, distending, and dull pain, while sharp pain includes at least tearing, cutting, stabbing, burning, and colic. Understanding pain types helps the recommendation system select consultation plans that are appropriate for different pain types.
[0119] Based on the pain area selected by the patient, the system automatically recommends motion examination plans in the horizontal, coronal and sagittal planes, and the patient follows the motion prompts (such as Figure 5 The system completes movement tests in the horizontal, coronal, and sagittal planes, and assigns a VAS score to each movement. The VAS score serves as the patient's personalized movement test feedback data. Through comprehensive analysis of movement test data in the three planes, the system can accurately locate the pain area and its impact on specific movements, quantify the degree of movement limitation of the patient, and formulate a personalized treatment plan based on this.
[0120] The multimodal data is then preprocessed, including: information encoding of the multimodal data, which includes mapping and standardizing the multimodal data according to a predefined data model; and then the information-encoded multimodal data is preprocessed by removing duplicate data, filling missing values, correcting format errors, and processing outliers.
[0121] The embedding layer is designed to convert discrete pain symptom information into a low-dimensional continuous vector representation and address data format consistency issues. The embedding layer captures the semantic relationships between pain symptom information by mapping these discrete variables into a continuous vector space. The structure of the embedding layer can be viewed as a lookup table that maps each discrete category to a vector of fixed dimension. Its core is a trainable weight matrix, in which each row represents the embedding vector for a category. The specific formula is as follows:
[0122] e i =W[x i ],i∈[0,N]
[0123] Wherein, N=4 represents the total number of pain disease information (i.e., the total number of modal types of multimodal data), x i Represents each pain condition information, e i represents the vector encoded by the embedding layer, and W represents the weight matrix;
[0124] Then the vector encoded by the embedding layer is input into the multimodal attention network in the intelligent analysis module;
[0125] In the previous step, the pain symptom information is separately encoded into a feature vector through the embedding layer. The data itself contains a large number of useless components and does not selectively focus on the data of the current modality. Therefore, the multimodal attention layer uses a multi-head self-attention mechanism to adaptively weight its own modal data, enhance the signal-to-noise ratio of the data features, and increase attention to the feature information that has a significant effect on the pain acupoints. The use of a multi-head design allows each attention mechanism to optimize the different feature parts of each modal data, thereby balancing the deviations that may be generated by the same attention mechanism and allowing the modal data to focus on features from multiple angles. The specific layer structure is as follows: Figure 6 As shown;
[0126] In the multimodal attention layer, the input is the encoding vector of each modality. First, the feature vector is multiplied by the weight matrices Wq, Wk, and Wv to obtain the query vector, key vector, and value vector. Their respective meanings are: query vector: represents the "query" initiated for the current information; key vector: indicates that there is a corresponding key in the input feature vector, and each key represents its related features; value vector: represents the actual data or features associated with each input, and the model ultimately uses these values to generate the output;
[0127] Next, the correlation score between the query vector and the key vector is calculated and used as a weighting coefficient. These scores are then passed through an activation function to obtain attention weights. This is then combined with the value vector to create a pain vector weighted by these weights. This allows the current modal data to focus on the pain input features relevant to the current query. This mechanism makes the attention mechanism highly effective in capturing important features, improving the model's ability to understand complex disease information, and thus providing more reliable data support for accurate diagnosis and the design of personalized treatment plans.
[0128] Then, the mutual attention layer is used to fuse the main complaint data. Its main function is to align and fuse the multimodal main complaint information, solve the differences in modal characteristics and distribution, and ensure that the output features are efficiently represented in a unified space. Its structure is as follows: Figure 7As shown in the figure, the mutual attention layer performs feature fusion on the multimodal high-dimensional vector based on the mutual attention mechanism. By calculating the similarity weights between each modality, it captures the mutual dependence between different modalities and generates weighted fusion features. Specifically, the input multimodal data includes the patient's chief complaint information (such as pain area, accompanying symptoms, pain intensity, pain duration, and action examination feedback data) and case information (such as previous pain history, treatment records, and medication use). These input data are first preprocessed and encoded to form multiple high-dimensional vectors. Then, through the mutual attention mechanism, the model can interact between different modalities, pay attention to and adjust the influence of each modality, and finally generate fusion features. These fusion features integrate all-round information of the patient's chief complaint information and medical history information.
[0129] The characteristics of pathological information features are that the data presents a discrete distribution and has a low feature dimension. In order to enhance the learning ability of the model, in this layer, the pathological information features are first projected into a high-dimensional space through the mapping layer; then, the multimodal pathological information is further fused through the mutual attention network, establishing associations between different data features and capturing their mutual influence and interaction; the core idea of the mutual attention mechanism is to guide the feature extraction of another modality by focusing on the important information in one modality, thus playing a vital role in multimodal fusion tasks or cross-modal learning tasks; finally, a multi-layer perceptron and nonlinear activation function are used to further nonlinearly combine and extract features from the high-order features extracted from different modal data, and a multi-layer perceptron is used to assign corresponding probabilities of meeting the treatment plan to 90 human consultation points constructed by professional physicians' experience, and output acupuncture points with probability values greater than the threshold to obtain a set of preliminary associated acupuncture points; this set does not integrate the human link association strength relationship information designed by professional physicians' experience, and requires further adjustment of the subsequent network based on expert experience;
[0130] The knowledge graph construction module is then used to construct an acupuncture point search knowledge graph based on the consultation experience of professional physicians. This graph links the pain location, pain nature, and movement test results with possible treatment options. For example, if a patient complains of low back pain and the pain is distributed on the coronal plane, the knowledge graph can point to a specific basic treatment plan for the lower back.
[0131] Specifically, in the knowledge graph construction module, the preset TCM meridian therapy experience data is imported into the graph database to form a knowledge graph for acupuncture points containing nodes such as pain, intensity, location, treatment plan, and their relationships, such as Figure 8As shown in the figure, based on this graph, a basic treatment plan with no personalized differences related to the current condition can be found; in addition, in the knowledge graph construction module, based on the human body link inference rules of Bai's physical therapy experience, according to the strength and weakness characteristics of the human body link association, an adjacency matrix diagram containing the mutual strength and weakness relationships between at least 90 diagnosis and treatment points is constructed, as shown in the figure. Figure 9 As shown in the figure, the nodes of the graph represent different acupuncture points, and the edges represent the direct relationship between acupuncture points. Following the acupuncture point constraints of Bai's therapy (each truncation has 2-3 diagnosis and treatment points in the horizontal, coronal and sagittal planes; the distal points are the wrist and ankle, with a total of 12 diagnosis and treatment points), the acupuncture point search logic of Bai's therapy is obtained. This module converts the experience of Bai's therapy into a mathematical language that can be understood by the network, forming knowledge rules.
[0132] Finally, the therapy plan recommendation module uses the associated acupuncture point set inferred by the intelligent analysis module, combined with the basic treatment plan and human link association diagram obtained by searching the knowledge graph constructed by the clinical experience of professional physicians, to calculate and generate a personalized treatment plan; in this embodiment, the graph attention network is used to optimize and correct the preliminary treatment plan to ensure that the treatment plan is suitable for the patient's specific physical condition, such as Figure 10 As shown in the figure, the associated acupoint feature points and the general physiotherapy plan obtained by the knowledge graph search are input into the graph attention layer. The graph attention layer is used to use the human body link relationship graph to model and semantically enhance the relationship between the general physiotherapy plan and the preliminary associated acupoints, and calculate the deep nonlinear features of the physiotherapy plan containing the human body link relationship. According to the three elements of the self-attention mechanism, the graph attention also contains three elements: Q, K, and V. Q is the feature vector of the central node, K is the feature vector of all neighboring nodes, and V is the new feature vector of the central node after the aggregation operation.
[0133] The role of optimization and calibration is to calibrate and optimize the inference results based on the actual experience of the physician when generating the final treatment plan. In the physiotherapy plan recommendation module, the graph attention layer and the multi-layer perceptron work together to complete the optimization and calibration of the treatment acupuncture points. First, the preliminary treatment plan is adjusted using the above-mentioned graph information and the graph attention layer. By introducing graph information and the graph attention network to capture the relationship and feature information between acupuncture points, the model can pay more attention to important neighboring nodes, thereby improving the model's ability to provide personalized treatment plans. The feature information of neighboring nodes is aggregated to obtain the feature information of the current node, making feature transfer more efficient, and can better capture the relationship between nodes and changes in feature information, thereby improving the generalization ability of the model. Then, the output of the graph attention layer is subjected to deep nonlinear feature extraction by the multi-layer perceptron to further capture the implicit pattern of acupuncture point combinations and optimize treatment decisions. By filtering invalid features, multi-layer mapping and feature calibration, the multi-layer perceptron outputs the final set of treatment acupuncture points. Overall, the graph attention layer is responsible for "relationship modeling and semantic enhancement", and the multi-layer perceptron is responsible for "feature optimization and personalized decision-making". The combination of the two ensures that the plan is both scientific and accurate and meets personalized needs.
[0134] Through a series of network architectures, the network can generate a personalized treatment plan based on the patient's main complaint information and the professional physician's experience in Traditional Chinese Medicine. This plan not only integrates the physician's experience to establish the relationship between the human meridian structure and the consultation points, but also uses the personalized plan provided by the professional physician consultation system as supervision to ensure that the network can provide a physical therapy plan that meets the patient's needs.
[0135] To further optimize the solution, this embodiment has designed a feedback update mechanism. After the patient receives treatment, the system collects efficacy data (such as the degree of pain relief, improvement in movement restrictions, etc.) to dynamically update the Bai's meridian acupoint experience knowledge map, continuously improving the accuracy of the acupoint search network model and the quality of the personalized treatment plan.
[0136] Specifically, the learning process of the model to recommend Bai's skin acupuncture consultation technology mainly consists of two parts: forward reasoning and feedback update;
[0137] Based on multimodal information input, the model gradually generates a forward reasoning process for personalized Bai's skin acupuncture consultation plans. First, the patient's chief complaint data (such as medical history information, pain location, pain type, and movement examination data) is input into the network; then the input information is encoded using the embedding layer, discrete pain symptom information is converted into a low-dimensional continuous vector representation, and a multimodal attention network is used to fuse multiple different pain symptom information encoding vectors to capture the semantic relationship between pain symptom information; then, based on the learned pain features and combined with the medical knowledge graph of Bai's skin acupuncture, the model can generate a basic treatment plan and adjust the basic treatment plan using graph information and the graph attention network, thereby improving the model's ability to provide personalized treatment plans; finally, the network can generate a personalized treatment plan that integrates the patient's chief complaint information and the professional physician's traditional Chinese medicine experience. At this point, the model completes the forward reasoning process and also learns the fusion relationship of different modal features and the deep association between symptoms and acupoints from the multimodal input data and the Bai's experience acupoint knowledge graph;
[0138] Feedback update is the core mechanism of this model to learn real-world medical consultation techniques and gradually improve them (e.g. Figure 11 After the model generates a consultation plan, the patient receives treatment according to the recommended acupoints and records the treatment effect through efficacy evaluation. These efficacy data, such as the degree of pain relief and recovery after treatment, will be input into the system as feedback information for further optimization of the model. After the feedback data enters the model, the calibration results are first deeply analyzed through the graph attention mechanism to extract more accurate feature information from the Bai's experience acupoint knowledge graph. At the same time, the model further adjusts the parameters in each module so that the model can more accurately align the relationship between the patient's symptoms and acupoints. Through this dynamic update process, the system gradually learns how to better deal with complex or special cases and improves its ability to handle diverse patient needs. In addition, feedback information is also used to update the Bai's experience knowledge graph. The system will re-evaluate the applicability of acupoints based on the actual feedback of treatment effects, strengthen the recording of effective acupoints, and mark points with poor efficacy and reduce their recommendation priority. Through the dynamic adjustment of the knowledge graph, the model can gradually form a more personalized and targeted consultation path.
[0139] Through forward reasoning and feedback updates, the system combines the patient's initial symptom information with feedback data to generate optimized treatment plans based on the patient's personalized symptom information and Bai's traditional Chinese medicine experience. At the same time, it continuously improves its own knowledge graph representation and enhances its understanding of the deep connection between symptoms and acupoints. The two form a closed-loop system. As patient samples accumulate, the model continues to learn from new cases and real-time treatment data, which not only improves the accuracy and personalization of consultation plans, but also optimizes its own knowledge representation and decision-making logic through continuous learning, forming a more dynamic and intelligent consultation recommendation system.
[0140] This system provides a more accurate and personalized treatment plan. While achieving personalized consultation, it improves consultation efficiency and reduces the burden on doctors. There is no need to wear fixed equipment to collect data during the movement inspection stage. The whole process is simple to operate and easy to promote. Non-professionals can also conduct independent pain consultation and rehabilitation management.
[0141] Example 3
[0142] This embodiment provides a verification experiment to verify the effectiveness of the recommendation system described in embodiment 1 or 2.
[0143] In the specific implementation process, this experiment recruited 128 patients with chronic pain for consultation. Among them, the experimental group consisted of 68 patients, who were consulted using the recommendation system provided in Example 1 or 2; the control group consisted of 60 patients, who were personally consulted by professional physicians. The statistical information of the patients' gender, age group, and pain location was as follows: Figure 12 and Figure 13 As shown in the figure, it can be seen that there is no statistically significant difference in gender, age group, and pain location between the experimental group and the control group. The degree of pain relief was calculated by VAS scores before and after treatment, and this was used as efficacy data to verify the accuracy and effectiveness of the consultation effect.
[0144] Equipment Introduction:
[0145] The core equipment of this experiment is the intelligent Chinese medicine meridian therapy instrument (such as Figure 14 It has a built-in Bai's experience knowledge graph and a multimodal consultation solution recommendation network, an integrated chronic pain management consultation system, and supports multi-point electrical stimulation therapy. The device adopts a non-invasive design, using electrodes to replace traditional acupuncture stimulation, and can accurately stimulate 10 independent treatment points. Each point provides 40 levels of current adjustment, covering multiple treatment modes such as sparse and dense waves, intermittent waves, and continuous waves. During the treatment process, the device can dynamically monitor the patient's physiological reactions and generate feedback data to support the subsequent optimization of the consultation solution. The device is used by professional physicians, and in the following comparative experiments, the same treatment device was used. The difference is that the experimental group used the built-in chronic pain management consultation system to recommend diagnosis and treatment plans, while the control group was given a treatment plan by a professional physician based on their own experience.
[0146] Operation process:
[0147] Patients first fill out personal information and medical records, including pain location, pain type, duration, and VAS score (Visual Analog Scale). The VAS score is an internationally accepted pain intensity quantification tool. Patients subjectively express their pain intensity by marking a straight line from "no pain" (0 points) to "severe pain" (10 points). Among them, 0 points indicates no pain, 1-3 points indicate mild pain, 4-6 points indicate moderate pain, and 7-10 points indicate severe pain. This score is used to quantify the changes in patients' pain before and after treatment, intuitively evaluate the treatment effect of the intelligent traditional Chinese medicine physiotherapy device, and verify the accuracy of the system-generated plan.
[0148] Patients in the experimental group were required to undergo movement checks and pain induction assessments. The intelligent TCM meridian therapy device accurately located the treatment points based on the patient's input data and the system-generated consultation plan, and implemented electrical stimulation therapy by applying electrodes. During the treatment, the device adjusted the treatment parameters in real time through multi-speed adjustment and multi-channel control to ensure personalized consultation results. After the treatment, the patient filled out the consultation record again to evaluate the pain relief, and the effect was analyzed in combination with the treatment data recorded by the instrument. In the control group, professional physicians personally diagnosed and implemented treatment for patients based on the consultation results. After the treatment, the patient filled out the consultation record again to evaluate the pain relief.
[0149] Statistical analysis of data after diagnosis and physical therapy of patients in the experimental group and the control group:
[0150] The results showed that among patients diagnosed and treated with the intelligent TCM meridian therapy device, 85.29% achieved a 40% to 100% pain relief, while 80% of patients who were personally consulted by professional physicians achieved a 40% to 100% relief. Among them, 29.41% of patients who used the device for consultation achieved a 70% to 100% relief, while this proportion was 18.33% of patients who were consulted by professional physicians, indicating that the consultation system designed by this patent is more effective than traditional physician consultation. In addition, the vast majority of patients reported that the pain area was significantly reduced, the range of motion was increased, and the treatment process was comfortable and easy to accept.
[0151] The statistical results are shown in Table 1:
[0152] Table 1 Comparison of the degree of relief after TCM physiotherapy in the experimental group and the control group
[0153]
[0154] Experimental results and discussion:
[0155] The experimental results (Table 1) show that the consultation plan generated by the intelligent TCM meridian therapy instrument based on the recommendation system in Example 1 or 2 is slightly better than the plan formulated by professional physicians in terms of efficacy, and the patient's pain relief and functional recovery effects are both as expected; the verification results prove that the recommendation system in Example 1 or 2 can effectively replace traditional doctor consultations, has wide promotion and application value, can provide safe and efficient treatment plans for patients with chronic pain, and significantly reduce the workload of physicians.
[0156] The same or similar reference numerals correspond to the same or similar components;
[0157] The terms used in the drawings to describe positional relationships are for illustrative purposes only and should not be construed as limiting this patent;
[0158] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.
Claims
1. An intelligent recommendation system for traditional Chinese medicine meridian therapy programs for chronic pain, characterized by: include: Multimodal data acquisition module, knowledge graph construction module, intelligent analysis module and physical therapy plan recommendation module; The multimodal data acquisition module is connected to the intelligent analysis module and the physical therapy plan recommendation module in sequence; the knowledge graph construction module is connected to the physical therapy plan recommendation module; The multimodal data acquisition module is used to collect and pre-process multimodal data of chronic pain patients, wherein the multimodal data includes at least: medical history information, pain location, pain type and personalized action examination feedback data; The knowledge graph construction module is used to construct an acupoint-finding knowledge graph and a human body link relationship graph based on preset TCM meridian therapy experience data; in the knowledge graph construction module, the preset TCM meridian therapy experience data is imported into the graph database to form an acupoint-finding knowledge graph; In the knowledge graph construction module, based on the preset TCM meridian therapy experience data and the strength characteristics of the human body link association, an adjacency matrix diagram containing the mutual strength relationships between at least 90 diagnosis and treatment points is constructed to obtain a human body link relationship diagram; in the human body link relationship diagram, the nodes of the diagram represent different acupuncture points, and the edges represent the direct associations between acupuncture points; The intelligent analysis module is pre-set with a multimodal attention network, which is used to perform feature extraction, feature fusion and preliminary acupoint prediction on the pre-processed multimodal data to generate a preliminary set of associated acupoints; In the intelligent analysis module, the structure of the multimodal attention network includes: an embedding layer, a multi-head self-attention layer, a mapping layer, a mutual attention layer, and a first multi-layer perceptron layer connected in sequence; The embedding layer is used to extract key features from the pre-processed multimodal data, map the discrete variables in the pre-processed multimodal data into a continuous vector space, and obtain a multimodal embedding vector; The multi-head self-attention layer is used to perform adaptive feature weighting on the multimodal embedding vector; The mapping layer is used to project the multimodal embedding vector after the multi-head self-attention layer into a higher-dimensional vector space to obtain a multimodal high-dimensional vector; The mutual attention layer performs feature fusion on the multimodal high-dimensional vector based on the mutual attention mechanism, captures the mutual dependence between different modalities by calculating the similarity weights between each modality, and generates fusion features after weighting; The first multi-layer perceptron layer is used to make preliminary probability predictions of acupoints based on the fusion features, and to group acupoints with the top predicted probabilities into a preliminary associated acupoint set for output; The therapy plan recommendation module is used to obtain a general therapy plan based on the acupuncture point search knowledge graph, and input the general therapy plan, the preliminary associated acupuncture point set and the human body link relationship diagram into a preset optimization calibration network, and finally generate a personalized Chinese medicine meridian therapy plan to complete intelligent recommendation.
2. The intelligent recommendation system for TCM meridian therapy for chronic pain according to claim 1, characterized in that: The multimodal data is obtained based on the patient's complaint, wherein the medical history information includes at least: information on the main complaint of pain area, information on the main complaint of accompanying pain area, information on the main complaint of pain intensity, information on the main complaint of pain duration, information on the main complaint of previous pain history, treatment records, and medication use; The pain site includes at least one of the head, neck, shoulder blade, chest, back, waist, pelvis and hip, leg, knee, ankle, foot, elbow, wrist or finger; The pain types include: dull pain or sharp pain, wherein dull pain includes at least soreness, distending pain and dull pain, and sharp pain includes at least tearing pain, cutting pain, stabbing pain, burning pain and colic; Based on the pain area selected by the patient, the patient completes the movement examination in the horizontal, coronal and sagittal planes according to the preset movement prompts, and performs a VAS score on each movement. The VAS score is used as the patient's personalized movement examination feedback data.
3. The intelligent recommendation system for TCM meridian therapy for chronic pain according to claim 1, characterized in that: The preprocessing includes: performing information encoding on the multimodal data, wherein the information encoding includes mapping and standardizing the multimodal data according to a predefined data model; The multimodal data after information encoding is preprocessed by removing duplicate data, filling missing values, correcting format errors and processing outliers.
4. The intelligent recommendation system for TCM meridian therapy for chronic pain according to claim 1, characterized in that: The preset TCM meridian therapy experience data is specifically medical record data related to TCM meridian therapy obtained from multiple sources.
5. The intelligent recommendation system for TCM meridian therapy for chronic pain according to claim 1, characterized in that: In the therapy plan recommendation module, the optimization calibration network includes: a graph attention layer and a second multi-layer perceptron layer connected in sequence; The universal physiotherapy plan, the preliminary associated acupoint set, and the human body link relationship graph are input into the graph attention layer. The graph attention layer is used to use the human body link relationship graph to perform relationship modeling and semantic enhancement on the universal physiotherapy plan and the preliminary associated acupoint set, and calculate the deep nonlinear features of the physiotherapy plan containing the human body link relationship; The second multi-layer perceptron layer is used to optimize and calibrate deep nonlinear features, capture the implicit combination patterns of acupuncture points, output the final acupuncture point prediction results, and save them as personalized Chinese medicine meridian therapy plans to complete intelligent recommendations.
6. The intelligent recommendation system for TCM meridian therapy for chronic pain according to any one of claims 1 to 5, characterized in that: The system further includes: a feedback update module; the feedback update module is connected to the knowledge graph construction module, the intelligent analysis module and the physical therapy plan recommendation module respectively; The feedback update module is used to update the cave-finding knowledge graph and fine-tune the multimodal attention network and the optimization calibration network.
7. The intelligent recommendation system for TCM meridian therapy for chronic pain according to claim 6, characterized in that: Chronic pain patients receive treatment based on personalized TCM meridian therapy plans and provide feedback on efficacy data to the feedback update module; The preset loss function is calculated based on the therapeutic effect data, the acupuncture knowledge graph is updated according to the preset loss function, and the multimodal attention network and the optimization calibration network are fine-tuned.
8. The intelligent recommendation system for TCM meridian therapy for chronic pain according to claim 7, characterized in that: The preset loss function is specifically: in, is the preset loss function value; It is a recovery index derived from efficacy data.