Knee arthralgia auxiliary traditional chinese medical treatment scheme generation method and system based on five-body hierarchy

Abstract

The application discloses a knee arthralgia auxiliary traditional Chinese medicine diagnosis and treatment scheme generation method and system based on five-body hierarchy. Firstly, the skin image, temperature, muscle strength curve, joint movement video, pulse and tongue image and X-ray image and other multi-source heterogeneous data are standardized collected according to the five-body arthralgia dimension of skin, muscle, tendon, vessel and bone; subsequently, the AI model is adopted to extract the features of the dimension data and quantitatively evaluate; then, the evaluation results of the dimensions are encoded and spliced into a fusion feature vector, the trained full connection neural network model is utilized for comprehensive classification, and the compound type of the five-body arthralgia is determined; finally, based on the compound type and grading result, the personalized diagnosis and treatment scheme including the traditional Chinese medicine treatment method, specific measures and rehabilitation tracking plan is automatically generated by matching the clinical rule base. The application realizes the objective quantification of the traditional Chinese medicine knee arthralgia diagnosis and the accurate identification of the arthralgia type, and can intelligently generate the personalized treatment scheme, thereby effectively improving the consistency and efficiency of the diagnosis and treatment.

Description

Technical Field

[0001] This invention belongs to the field of TCM intelligent diagnosis, rehabilitation assessment and precision medicine technology, specifically involving a method and system for generating auxiliary TCM diagnosis and treatment plans for knee arthritis based on the five body levels. Background Technology

[0002] Knee arthritis (corresponding to osteoarthritis, synovitis, etc. in Western medicine) is a common degenerative joint disease among middle-aged and elderly people, and there are several key challenges in the traditional Chinese medicine diagnosis and treatment practice: I. Diagnosis and grading are highly subjective. Traditional diagnosis and treatment mainly rely on the physician's experience to qualitatively identify the symptoms of the five bodily systems: skin, flesh, tendons, blood vessels, and bones. The lack of standardized quantitative diagnostic indicators and disease grading system leads to insufficient consistency in diagnosis among different physicians and limited accuracy of identification.

[0003] II. High difficulty in integrating and analyzing multi-dimensional data Knee pain involves multi-source heterogeneous data (such as image data: photos of skin appearance, medical images; time-series data: muscle strength change curves, motion video recordings; text data: descriptions of syndrome differentiation, medical history, etc.). Currently, there is a lack of standardized multimodal fusion methods suitable for clinical diagnosis and classification, making it difficult to support accurate disease assessment.

[0004] III. Low efficiency in matching rehabilitation programs and tracking treatment effectiveness. The correlation between complex types of arthralgia (a traditional Chinese medicine concept) and TCM treatments and rehabilitation interventions largely relies on human experience, making it difficult to quickly generate individualized plans based on disease severity levels. Furthermore, the lack of effective dynamic monitoring methods for the rehabilitation process hinders the scientific assessment of treatment efficacy and the optimization of treatment plans. Summary of the Invention

[0005] In view of this, the first aspect of the present invention provides a method for generating a traditional Chinese medicine diagnosis and treatment plan for knee arthritis based on the five-body level, comprising the following steps: Step S1: Standardized collection of multi-source data at five body levels: Standardized data is collected for each of the five body pain dimensions: skin, muscle, tendon, blood vessels, and bone. Step S2: AI Feature Extraction and Evaluation of Single Physical Pain Dimension: Based on the 0-corresponding data collected in Step S1, an appropriate AI model or algorithm is used to extract and evaluate features for each physical pain dimension, and output the diagnostic results and classification of each physical pain. Step S3: Multi-source heterogeneous data fusion and comprehensive classification of arthralgia: The arthralgia assessment results output in step S2 are numerically encoded and concatenated to form a fusion feature vector; the fusion feature vector is input into a trained fully connected neural network classification model to obtain the probability of the presence or absence of the five arthralgias; based on the probability of the probability of the probability, the complex type of the five arthralgias of the patient is determined. Step S4: Generation of Personalized TCM Treatment Plan: Based on the diagnostic results of the five types of arthralgia and the grade of each arthralgia obtained in Step S3, and combined with the patient's basic information, a personalized treatment plan is generated through rule matching.

[0006] The standardized data includes: Knee joint skin images and skin surface temperature data were collected as dermatopathic data; Muscle strength-time curve data from isokinetic muscle strength tests of the knee joint were collected as data for muscle paralysis. Videos of knee flexion, extension, and rotation movements, along with coordinate data of key joint points, were collected as muscle and joint data. Pulse parameters, tongue images, and structured medical history information were collected as pulse obstruction data. Knee X-ray images and associated pain scores were collected as bone pain data.

[0007] Based on the above scheme, step S2 specifically includes: For dermatophyte data, a medical-grade lightweight CNN model is used to extract skin image features, and combined with temperature data for fusion analysis to output dermatophyte diagnosis and grading; Based on the data on muscular dystrophy, the rate of muscle strength reduction is calculated using the peak torque algorithm and the ratio of flexor to extensor torque, and the diagnosis and grading of muscular dystrophy are output by combining statistical models. For muscle pain data, an RNN model is used to process the coordinate sequence of key joint points, extract kinematic features, and output muscle pain diagnosis and grading. For pulse obstruction data, CNN is used to extract tongue surface image features, and RNN is used to process pulse time series data, BiLSTM model is used to process the fusion sequence of tongue surface features and syndrome differentiation information, and after fusing structured consultation information, the pulse obstruction assessment results are output. For osteoarthritis data, based on knee X-ray images, the joint space is measured using the KL grading standard and edge detection algorithm to output osteoarthritis diagnosis and grading.

[0008] Based on the above scheme, in step S3, the standardized encoding of each body pain assessment result is concatenated into a 128-dimensional fusion feature vector.

[0009] The feature vector consists of the following features: The skin texture features are composed of color labels, temperature statistical features, and skin texture features extracted by CNN; The characteristics of muscular paralysis are composed of the rate of decrease in peak torque, the ratio of flexor to extensor torque, stability abnormality labels, and temporal characteristics of muscle strength curves. The characteristics of muscle paralysis are composed of joint angle deviation, muscle and knee elasticity labels, movement pattern disorder, and key point sequence features extracted by RNN. The characteristics of pulse obstruction are composed of blood circulation labels, pain type labels, pulse characteristics, tongue surface CNN characteristics, and pulse temporal characteristics; The characteristics of osteoarthritis are composed of KL grade, joint space width, VAS pain score and image segmentation features.

[0010] Based on the above scheme, the standardized coding includes skin dimensional coding, and the generated skin features are 24-dimensional. The features consist of: a 1-dimensional color label generated based on the skin image, a 3-dimensional temperature statistical feature calculated based on the temperature values ​​of at least three standard sites, and a 20-dimensional texture feature vector extracted from the skin image by a convolutional neural network.

[0011] The standardized coding includes muscle spasm dimension coding, and the generated muscle spasm features are 28-dimensional, which consist of: a 1-dimensional peak torque reduction rate, a 1-dimensional flexion-extension torque ratio, a 1-dimensional stability anomaly label calculated based on isokinetic muscle strength test data, and a 25-dimensional temporal feature vector extracted from the muscle strength-time curve by a long short-term memory network.

[0012] The standardized coding includes muscle and tendon dimension coding, and the generated muscle and tendon features are 24-dimensional. The features consist of: a 1-dimensional joint angle deviation value calculated based on the coordinate sequence of key joint points, a 1-dimensional muscle and knee elasticity label, a 1-dimensional motion pattern disorder degree, and a 21-dimensional motion temporal feature vector extracted from the coordinate sequence by a recurrent neural network.

[0013] The standardized coding includes pulse obstruction dimension coding, and the generated pulse obstruction features are 24-dimensional, which consist of: 1-dimensional blood vessel operation status label, 1-dimensional pain type label, 3-dimensional pulse feature value, 6-dimensional tongue surface feature vector extracted from tongue surface image by convolutional neural network, and 13-dimensional pulse time sequence feature vector extracted from pulse time sequence data by recurrent neural network.

[0014] The standardized coding includes bone pain dimension coding, and the generated bone pain features are 28-dimensional, which consists of: 1-dimensional KL grade value, 1-dimensional joint space width measurement value, 1-dimensional VAS pain score value, and 25-dimensional image structure feature vector extracted from knee X-ray images by the image segmentation model.

[0015] Based on the above scheme, the comprehensive classification using a classification model specifically involves: inputting the fused feature vector into a fully connected neural network classification model; wherein the classification model has an input layer, a first hidden layer, a second hidden layer, and an output layer connected in sequence; the number of neurons in the input layer is the same as the dimension of the fused feature vector, the first hidden layer has 64 neurons and uses the ReLU activation function, the second hidden layer has 32 neurons and uses the ReLU activation function, and the output layer has 5 neurons and uses the Sigmoid activation function, respectively outputting the existence probability values ​​corresponding to the five dimensions of physical pain: skin, muscle, tendon, vein, and bone.

[0016] Based on the above scheme, the method further includes a step of optimizing scheme generation based on association rule mining: using the Apriori algorithm to mine the association between the combination of physical pain features and traditional Chinese medicine treatment methods in historical data; wherein, the minimum support is set to 5% and the minimum confidence is set to 70%; and the strong association rules that meet the conditions are stored in the clinical rule base for matching and recommendation when generating schemes.

[0017] Based on the above scheme, the determination of the five types of physical pain is specifically included as follows: comparing the probability values ​​of each physical pain output by the classification model with a preset determination threshold; if the probability value of a certain physical pain is greater than or equal to the determination threshold, then the physical pain is determined to exist; combining all the physical pain types that are determined to exist to obtain the determination result of the composite type; wherein, the preset determination threshold is 0.6.

[0018] Based on the above scheme, the automatic generation of personalized TCM diagnosis and rehabilitation plans specifically includes a rule matching step: taking the composite type determination result and the information on the grade of each type of knee pain as input, and calling a clinical-level TCM treatment method association library for matching; wherein, the association library is constructed in the following way: using TCM clinical guidelines as initial rules, and performing association verification and optimization based on actual treatment data and efficacy feedback of more than 3,000 historical clinical cases of knee pain, forming mapping rules between the combination of knee pain characteristics, grade and TCM treatment methods, and intervention intensity.

[0019] Based on the above approach, the personalized diagnosis and rehabilitation plan includes the following structured output: A standardized diagnostic and identification report shall include at least the following fields: diagnostic results for the five types of arthralgia, classification, core features, and diagnostic basis based on the multi-source heterogeneous data; Personalized treatment sub-plans include at least: traditional Chinese medicine treatments with intervention intensity appropriate to the type and grade of physical pain, as well as one or more combinations of specific Chinese herbal prescriptions, acupuncture points, and physical therapy methods; A rehabilitation follow-up plan should include at least the following: standardized rehabilitation training movements, a pre-set follow-up examination cycle, and quantified indicators for evaluating the effectiveness of the treatment.

[0020] Secondly, a system for generating auxiliary TCM diagnosis and treatment plans for knee arthritis based on the five-body hierarchy is provided, including: The data acquisition module is used to collect standardized data for the five dimensions of physical pain: skin, flesh, tendons, blood vessels, and bones. A single-dimensional assessment module, connected to the data acquisition module, is used to extract and assess features of each physical pain dimension based on the acquired corresponding data and using an appropriate AI model or algorithm, and output the diagnostic results and grading of each physical pain. The fusion classification module, connected to the single-dimensional assessment module, is used to numerically encode the assessment results of each physical ailment and concatenate them to form a fusion feature vector. The fusion feature vector is then input into a trained fully connected neural network classification model to obtain the probability of the presence or absence of the five physical ailments. Based on the probability of the probability of the probability, the patient's five physical ailment complex type is determined. The treatment plan generation module, connected to the fusion classification module, is used to generate a personalized treatment plan that includes traditional Chinese medicine treatment methods, specific measures, and rehabilitation follow-up plans based on the diagnostic results of the five types of paralysis and the grade of each paralysis, combined with the patient's basic information, through rule matching.

[0021] The beneficial effects of this invention are: (1) This invention achieves objectification and quantification in the diagnosis of knee arthralgia in Traditional Chinese Medicine (TCM), improving consistency and repeatability: Based on the TCM theory of "skin, flesh, tendons, vessels, and bones" for arthralgia, this invention designs a complete multi-source heterogeneous data acquisition and quantitative evaluation system. Through various technical means such as high-definition images, infrared thermometry, isokinetic muscle strength, motion capture, pulse diagnosis and tongue diagnosis, and medical imaging, the traditional qualitative diagnosis that relies on the subjective experience of physicians is transformed into quantitative features based on objective data (such as temperature statistics, peak torque reduction rate, joint angle deviation, pulse characteristic values, joint space width, etc.). This fundamentally solves the core pain points of traditional diagnosis and treatment being highly subjective and having poor consistency in diagnoses among different physicians, making the diagnostic process verifiable, and the results quantifiable, comparable, and traceable.

[0022] (2) A multi-dimensional data fusion analysis model was constructed, enabling accurate identification of complex arthralgia types: This invention achieves deep fusion and comprehensive analysis of information from five dimensions—skin, muscle, tendon, vein, and bone—through innovative feature encoding rules (such as encoding the five body features into a 128-dimensional fusion vector) and a specially designed fully connected neural network classification model. The system can automatically determine single or compound arthralgia types (such as "skin arthralgia + muscle arthralgia + bone arthralgia") and perform accurate classification based on probability thresholds (such as 0.6), solving the problem of the difficulty in comprehensively weighing multiple arthralgia signs manually, and providing a reliable classification basis for precision treatment.

[0023] (3) It provides data-driven intelligent generation capabilities for personalized treatment plans, improving treatment adaptability and efficiency: This invention deeply integrates artificial intelligence technology with TCM clinical knowledge. On the one hand, it uses the Apriori algorithm to mine strong correlation rules (such as support of 5% and confidence of 70%) between "physical arthralgia characteristics and TCM treatment methods" in massive clinical data, and constructs and continuously optimizes an intelligent rule base. On the other hand, the system can automatically match and generate structured personalized plans based on precise complex arthralgia types, specific grades of each type of arthralgia, and individual patient information, including TCM prescriptions, acupuncture points, physiotherapy methods, and rehabilitation plans. This changes the efficiency bottleneck of traditionally relying entirely on physicians' personal memory and experience to match plans, and realizes rapid, standardized, and personalized output of treatment plans. Attached Figure Description

[0024] The present invention includes the following figures: Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0026] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0027] like Figure 1 , 2 As shown, the specific implementation method of the present invention for generating a traditional Chinese medicine diagnosis and treatment plan for knee arthritis based on the five-body level includes the following technical steps: Step 1: Standardized collection of multi-source data at five body levels.

[0028] For the five dimensions of physical pain—skin, flesh, tendons, blood vessels, and bones—corresponding standardized data (including equipment and data format requirements) were collected: 1. Data collection for skin paralysis: Equipment: High-definition digital camera (resolution ≥ 1080P), infrared temperature sensor (accuracy ± 0.1℃); Procedure: Take anteroposterior and lateral images of the knee joint skin (under a uniform environment with a light intensity of 500 lux); collect temperature values ​​at three points on the skin surface of the knee joint (superior border of the patella, medial collateral ligament, and lateral collateral ligament); Data format: Images are in JPG format, and temperatures are in CSV format (including location, value, and acquisition time).

[0029] 2. Data collection on paralysis: Equipment: Isokinetic muscle strength testing system (such as Biodex System 4); Procedure: The patient sits with the knee joint fixed in a 90° flexion position. Perform flexion and extension movements 5 times each at angular velocities of 60° / s and 120° / s. Data format: Output muscle strength-time curve (CSV format, including angular velocity, peak torque, and flexor-extensor torque ratio).

[0030] 3. Collection of muscle and tendon pain data: Equipment: Motion capture camera (such as Kinect 2.0), tripod (fixed height 1.5m); Procedure: Record a video of the patient's knee joint flexion-extension-rotation movements (10 seconds in length, 30fps in frame rate). Data format: MP4 video format, synchronous output of joint key point coordinates (CSV format).

[0031] 4. Pulse pain data collection: Equipment: Four diagnostic instruments (pulse diagnosis + tongue diagnosis + questioning); Procedure: Collect parameters such as pulse rate, pulse amplitude, and pulse shape, as well as tongue images, according to the standard procedure. Record diagnostic information such as "blood circulation status (stasis / smooth flow)", "pain characteristics (stabbing / ache, etc.)" and "duration of illness" through medical history taking. Data format: Pulse data is in CSV format (including timestamp, pulse rate, peak pulse amplitude, pulse shape characteristic value, and comparison value of three pulse position parameters), tongue image is in JPG format (with acquisition parameters), and consultation information is in structured JSON format (including field validation rules to ensure data integrity).

[0032] 5. Bone pain data collection: Equipment: Knee X-ray machine (DR equipment); Procedure: Take anteroposterior and lateral X-ray images of the knee joint according to clinical standards. The radiologist marks the joint space measurement points and records the patient's VAS pain score (0-10 points) and imaging position parameters simultaneously. Data format: DICOM standard format image files (including patient basic information, imaging parameters, and measurement annotations), and associated data in CSV format (including joint space measurement values, VAS scores, and measuring physician information).

[0033] Step 2: AI feature extraction and evaluation for a single physical paralysis dimension.

[0034] For different data types, appropriate AI models are used to complete feature extraction and paralysis assessment: 1. Dermatoparesis assessment (CNN model): Input: Knee joint skin image, 3-site temperature data; Model: Medical-grade lightweight CNN (Modified MobileNetV2, input size 224×224), training dataset consists of 1000 clinically diagnosed knee arthritis patients' skin paralysis images + temperature labels (including healthy control samples). Processing steps: CNN extracts diagnostic features such as skin color (yellow / red / white, etc.) and texture; temperature data is input into a fully connected layer after environmental calibration and normalization; the fusion outputs the skin paralysis diagnosis result (such as "skin paralysis (yellow / red indicates heat)") and the grade (mild / moderate / severe, determined based on temperature deviation and degree of color abnormality).

[0035] 2. Physical paralysis assessment (peak moment algorithm + statistical model): Input: Peak torque and flexor-extensor torque ratio data from isokinetic muscle strength testing; Algorithm: Peak torque assessment algorithm (calculates the average peak torque at an angular velocity of 60° / s and compares it with the threshold of healthy individuals); Processing steps: Calculate the peak torque reduction rate = (health threshold - patient calibration value) / health threshold × 100%; Combine the flexor-extensor torque ratio to determine the diagnosis of paralysis (e.g., "paralysis (weakened muscle strength + decreased joint stability)") and its grade: mild (reduction rate 10%-20%), moderate (20%-30%), severe (>30%), where a reduction rate >20% and a flexor-extensor torque ratio >1.5 are the diagnostic criteria for moderate and above.

[0036] 3. Muscle pain assessment (RNN model): Input: Keypoint coordinate sequence of a knee joint motion video; Model: RNN (LSTM layers × 2, hidden layer dimension 64); Processing steps: RNN extracts diagnostic features such as joint angle deviation, movement pattern disorder, and muscle and knee elasticity; outputs the diagnosis results of muscle pain (such as "muscle pain (limited joint angle + movement pattern disorder)") and the grade (mild: angle deviation 10%-20%; moderate: 20%-30%; severe: >30%).

[0037] 4. Pulse obstruction assessment (RNN+BiLSTM model): Input: Tongue surface image features (extracted via CNN), pulse time series data (pulse rate, pulse amplitude, pulse shape feature sequence), and structured diagnostic information; Model: RNN (processes pulse time series data, LSTM layer × 2) + BiLSTM (processes tongue surface features and syndrome differentiation information fusion sequence, hidden layer dimension 80). Processing steps: CNN extracts tongue color and coating features; RNN captures the temporal variation pattern of pulse (such as abnormal pulse amplitude fluctuation and irregular pulse rate); integrates tongue features, pulse features and syndrome differentiation information; outputs pulse obstruction assessment results (such as "pulse obstruction (blood stasis + hesitant pulse + stabbing pain)" or "pulse obstruction (poor blood circulation + wiry pulse + distending pain)").

[0038] 5. Bone pain assessment (KL grading + image processing algorithm): Input: Knee joint X-ray DICOM image; Algorithm: KL grading (severity grading of osteoarthritis) + edge detection algorithm (such as Canny operator); Processing steps: Accurately measure the joint space width using image processing algorithms, combined with KL grading; output the diagnosis result of osteoarthritis (such as "Osteoarthritis (K-L2 grade + joint space narrowing)") and grading (mild: K-L1 grade; moderate: K-L2-3 grade; severe: K-L4 grade).

[0039] Step 3: Multi-source heterogeneous data fusion and comprehensive classification of paralysis.

[0040] 1. Feature splicing: The assessment results of the five types of paralysis were numerically encoded in modules and then concatenated into a 128-dimensional feature vector. The dimensional allocation and encoding rules for each type of paralysis are as follows: Dimensional Allocation and Coding Rules for Various Types of Bias Skin paralysis Color label (1) + Temperature statistical features (mean, variance, difference from healthy baseline) (3) + Skin texture features extracted by CNN (20) Total 24 dimensions: Color "Hot" → 1, "Cold" → 0; Temperature mean normalized to [0,1]; 20-dimensional texture features extracted by CNN (MobileNetV2). paralysis Peak torque reduction rate (1) + flexor-extensor torque ratio (1) + stability abnormality label (1) + temporal characteristics of muscle strength curve (25) Total 28 dimensions: Peak moment reduction rate = (healthy threshold - patient value) / healthy threshold, normalized to [0,1]; Abnormal stability → 1, normal → 0; 25-dimensional temporal features extracted by LSTM. Muscle paralysis Joint angle deviation (1) + tendon-knee elasticity label (1) + movement pattern disorder (1) + key point sequence features extracted by RNN (21) Total 24 dimensions: Angle deviation = (healthy angle - patient angle) / healthy angle, normalized to [0,1]; Knee elasticity present → 1, absent → 0; 21-dimensional sequence features output by RNN. Pulse paralysis Blood circulation label (1) + Pain type label (1) + Pulse characteristics (normalized pulse rate, peak pulse amplitude, pulse shape characteristic value) (3) + Tongue surface CNN features (6) + Pulse temporal characteristics (13) Total 24 dimensions: Blood stasis → 1, smooth flow → 0.5, fullness → 0; Pain: stabbing pain → 1, soreness → 0.5, distending pain → 0; Pulse rate and pulse amplitude peak values ​​are normalized to [0,1], pulse shape feature values ​​are encoded as wiry pulse → 0.8, hesitant pulse → 0.6, slippery pulse → 0.4; 6-dimensional tongue surface features are extracted by CNN, and 13-dimensional pulse temporal features are output by RNN. Bone paralysis KL classification (1) + joint space width (1) + VAS pain score (1) + image segmentation features (25) Total 28 dimensions: KL classification (levels 1-4) normalized to [0,1]; joint space width normalized to [0,1]; 25-dimensional image features extracted after U-Net segmentation. Finally, the features of the above 5 types of paralysis (24+28+24+24+28) are spliced ​​together into a 128-dimensional fusion feature vector.

[0041] 2. The specific structure and training of a fully connected neural network classification model This model is a multi-label classification model (outputting a "presence / absence" judgment for 5 types of paralysis), and its structure and training parameters are as follows: Network structure: Input layer (128-dimensional) → Hidden layer 1 (64 neurons, ReLU activation function) → Hidden layer 2 (32 neurons, ReLU activation function) → Output layer (5 neurons, Sigmoid activation function, corresponding to 5 probabilities of the presence of paralysis) Training configuration: Dataset: 5000 patients with knee pain, consisting of 128 fused features + 5 individual pain labels (presence → 1 / absence → 0); Loss function: Binary Cross-Entropy (BCE), suitable for multi-label classification scenarios; Optimizer: Adam optimizer, initial learning rate 0.001, decreasing during training as the validation set loss increases (decay coefficient 0.5); Training parameters: BatchSize=32, Epoch=50, using "early stopping" (training terminates if the validation set loss does not decrease for 5 consecutive epochs); Model output: probability values ​​in 5 dimensions (range [0,1]), used to determine whether each body has paralysis.

[0042] 3. Specific implementation of association rule mining (Apriori algorithm) The steps for mining association rules between "fusion feature terms → traditional Chinese medicine treatment methods" using the Apriori algorithm are as follows: (1) Transaction set construction: Discretize the fusion features of each patient into "items" (such as "skin numbness and fever", "muscle numbness and weakness", "pulse numbness and blood stasis"), and form transactions with the corresponding TCM treatments (such as "activating blood circulation and removing blood stasis" and "draining heat and dispelling cold"). Example: Item 1: {Skin numbness and heat, muscle numbness and weakness, blood stasis and stagnation} → Treatment: Promote blood circulation and remove blood stasis + dispel heat and cold.

[0043] (2) Parameter settings: Minimum support: 5% (i.e., at least 250 out of 5000 cases); Minimum confidence level: 70% (i.e., the credibility of the rule is ≥70%).

[0044] (3) Rule mining results: Output strong association rules, example: Rule 1: {Skin numbness and heat ∧ Vessel numbness and stagnation} → Promote blood circulation and remove blood stasis (Support 6%, Confidence 85%) Rule 2: {muscle weakness and paralysis ∧ muscle paralysis and abnormal angle} → improve muscle circulation (support 8%, confidence 78%).

[0045] 4. Logic for determining the complex type of five-body pain Based on the output probability of the fully connected model, a decision threshold (0.6) is set: if the output probability corresponding to a certain physical ailment is ≥0.6, then the physical ailment is determined to "exist"; combining all "existing" physical ailment types yields a composite type, as shown in the example: Probability of skin paralysis = 0.8 (≥0.6), probability of muscle paralysis = 0.7 (≥0.6), probability of tendon paralysis = 0.65 (≥0.6), probability of pulse paralysis = 0.72 (≥0.6), probability of bone paralysis = 0.61 (≥0.6) → Composite type: skin paralysis + muscle paralysis + tendon paralysis + pulse paralysis + bone paralysis.

[0046] Step 4: Generation of Personalized TCM Treatment Plans Input: Diagnosis and identification results of complex arthralgia (including the grade of arthralgia in each body), patient's basic information (age, gender, VAS pain score, course of disease, past medical history); Rule matching: Call the clinical-level TCM treatment method association library (built based on the clinical guidelines of "Traditional Chinese Medicine Orthopedics and Traumatology" + 3000 clinical cases for verification), and combine it with the graded and appropriate intervention intensity; Output: Standardized diagnostic report: including diagnostic results, grading, core characteristics, diagnostic basis (data support), and determination of complex arthralgia type; Personalized treatment plan: treatment method (intervention intensity appropriate to the grade) + specific measures (traditional Chinese medicine prescriptions, acupuncture points, physiotherapy); Rehabilitation follow-up plan: including rehabilitation movement specifications such as muscle strength training and joint range of motion training, follow-up examination cycle, and efficacy evaluation indicators (such as peak torque improvement rate, joint space changes, and VAS score reduction).

[0047] Based on the above approach, taking a 55-year-old female patient with knee pain as an example, the specific implementation method is as follows: Patient basic information Name: Zhang XX; Age: 55; Gender: Female; Chief complaint: Knee pain for 3 months, aggravated by flexion and extension, accompanied by skin redness and muscle soreness; No history of bone and joint surgery; VAS pain score: 6.

[0048] Step 1: Data Collection Results Skin paralysis: Skin image (yellowish-red in color), patellar superior margin temperature 37.8℃ (1.2℃ higher than healthy baseline); Muscular paralysis: Peak torque at 60° / s angular velocity 120 N·m (health threshold 150 N·m, reduction rate 20%), flexor-extensor torque ratio 1.6; Muscle stiffness: The maximum flexion and extension angle of the knee joint is 100° (healthy value is 120°, deviation is 16.7%), indicating "muscle and knee elasticity". Pulse obstruction: Tongue image (red tongue, thin yellow coating), pulse data (pulse rate 68 beats / min, peak pulse amplitude 0.35V, pulse shape characteristic value 0.6 (hesitant pulse), the parameters of the three pulse positions of Cun, Guan and Chi are "weak Cun and hesitant Guan"), syndrome differentiation information (blood circulation "stagnation", pain characteristics "stabbing pain"). Bone pain: X-ray KL grade 2, joint space 2.0mm (healthy value 3.0mm, deviation 33.3%).

[0049] Step 2: Results of Individual Physical Pain Assessment Skin numbness: CNN outputs "Skin numbness (yellowish-red indicates heat), moderate"; Paralysis: The peak moment algorithm outputs "paralysis (weakened muscle strength + decreased joint stability), moderate"; Muscle paralysis: The RNN outputs "Muscle paralysis (limited joint angle + disordered movement pattern), moderate"; Pulse obstruction: RNN+BiLSTM outputs "Pulse obstruction (blood stasis + hesitant pulse + stabbing pain)"; Bone pain: Image processing + KL grading output "Bone pain (K-L2 grade + joint space narrowing), moderate".

[0050] Step 3: Classification of Paralysis Syndromes After integrating the features into the diagnostic model, the output is "Five-Body Complex Bi (Moderate)"; the core treatment method is derived through clinical-grade Apriori algorithm: heat dissipation and cold dispersing (moderate intervention) + spasm relaxation and pain relief (moderate intervention) + blood circulation promotion and stasis removal (moderate intervention) + muscle and tendon mobilization and collaterals activation (moderate intervention) + tendon and bone setting (moderate intervention).

[0051] Step 4: Personalized Treatment Plan 1. Core conclusion of the diagnostic report: The patient was diagnosed with moderate five-body complex arthralgia. The core pathological features were hot skin, weak muscles, stagnant tendons, blood stasis, and bone damage. The diagnostic basis was abnormal skin temperature, 20% decrease in muscle strength, 16.7% deviation in joint angle, red tongue with stinging pain, and K-L2 grade narrowing of joint space. 2. Personalized treatment plans: Traditional Chinese medicine prescription: Modified Simiao Powder (Atractylodes lancea 10g, Phellodendron chinense 10g, Coix lacryma-jobi 30g, Achyranthes bidentata 15g, Salvia miltiorrhiza 15g), 1 dose per day, for 4 consecutive weeks; Acupuncture points: Dubi (ST35), Neixiyan (ST36), Yanglingquan (GB34), Xuehai (SP10). Retain needles for 20 minutes, twice a week, for a total of 8 sessions. Physiotherapy: Traditional Chinese medicine fumigation and washing (once a day, 20 minutes each time) + medical isokinetic muscle strength training (3 times a week, with an initial load of 60% of the healthy threshold, gradually increasing). 3. Rehabilitation follow-up plan: A follow-up examination will be conducted after 4 weeks. The efficacy evaluation indicators are: VAS score ≤ 4 points, peak torque improvement ≥ 10%, and joint flexion-extension angle ≥ 110°. The plan will be adjusted according to the follow-up examination results.

[0052] Based on the same inventive concept, in another specific embodiment, a system for generating auxiliary TCM diagnosis and treatment plans for knee arthritis based on the five-body hierarchy includes: The data acquisition module is used to collect standardized data for the five dimensions of physical pain: skin, flesh, tendons, blood vessels, and bones. A single-dimensional assessment module, connected to the data acquisition module, is used to extract and assess features of each physical pain dimension based on the acquired corresponding data and using an appropriate AI model or algorithm, and output the diagnostic results and grading of each physical pain. The fusion classification module, connected to the single-dimensional assessment module, is used to numerically encode the assessment results of each physical ailment and concatenate them to form a fusion feature vector. The fusion feature vector is then input into a trained fully connected neural network classification model to obtain the probability of the presence or absence of the five physical ailments. Based on the probability of the probability of the probability, the patient's five physical ailment complex type is determined. The treatment plan generation module, connected to the fusion classification module, is used to generate a personalized treatment plan that includes traditional Chinese medicine treatment methods, specific measures, and rehabilitation follow-up plans based on the diagnostic results of the five types of paralysis and the grade of each paralysis, combined with the patient's basic information, through rule matching.

[0053] It should be noted that any process or method description in the embodiments can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which the embodiments of the invention pertain.

[0054] It should be noted that the logic and / or steps in the embodiments, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0055] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0056] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0057] Furthermore, in the embodiments of the present invention, the functional modules can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The aforementioned storage medium can be a read-only memory, a hard disk, or an optical disk, etc.

[0058] The above embodiments have provided a detailed description of the technical solutions of the present invention. Obviously, the present invention is not limited to the described embodiments. Based on the embodiments of the present invention, those skilled in the art can make various changes, but any changes that are equivalent or similar to the present invention fall within the scope of protection of the present invention. Contents not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for generating an auxiliary TCM diagnosis and treatment plan for knee arthritis based on the five-body level, characterized in that, Includes the following steps: Step S1: Standardized collection of multi-source data at five body levels: Standardized data is collected for each of the five body pain dimensions: skin, muscle, tendon, blood vessels, and bone. Step S2: AI Feature Extraction and Evaluation of Single Physical Pain Dimension: Based on the corresponding data collected in Step S1, an appropriate AI model or algorithm is used to extract and evaluate features for each physical pain dimension, and output the diagnostic results and classification of each physical pain. Step S3: Multi-source heterogeneous data fusion and comprehensive classification of arthralgia: The arthralgia assessment results output in step S2 are numerically encoded and concatenated to form a fusion feature vector; the fusion feature vector is input into a trained fully connected neural network classification model to obtain the probability of the presence or absence of the five arthralgias; based on the probability of the probability of the probability, the complex type of the five arthralgias of the patient is determined. Step S4: Generation of personalized TCM treatment plan: Based on the diagnosis results of the five types of arthralgia and the grade of each arthralgia obtained in step S3, combined with the patient's basic information, a personalized treatment plan is generated through rule matching. The standardized data includes: Knee joint skin images and skin surface temperature data were collected as dermatopathic data; Muscle strength-time curve data from isokinetic muscle strength tests of the knee joint were collected as data for muscle paralysis. Videos of knee flexion, extension, and rotation movements, along with coordinate data of key joint points, were collected as muscle and joint data. Pulse parameters, tongue images, and structured medical history information were collected as pulse obstruction data. Knee X-ray images and associated pain scores were collected as bone pain data.

2. The method according to claim 1, characterized in that, Step S2 specifically includes: For dermatophyte data, a medical-grade lightweight CNN model is used to extract skin image features, and combined with temperature data for fusion analysis to output dermatophyte diagnosis and grading; Based on the data on muscular dystrophy, the rate of muscle strength reduction is calculated using the peak torque algorithm and the ratio of flexor to extensor torque, and the diagnosis and grading of muscular dystrophy are output by combining statistical models. For muscle pain data, an RNN model is used to process the coordinate sequence of key joint points, extract kinematic features, and output muscle pain diagnosis and grading. For pulse obstruction data, CNN is used to extract tongue surface image features, and RNN is used to process pulse time series data, BiLSTM model is used to process the fusion sequence of tongue surface features and syndrome differentiation information, and after fusing structured consultation information, the pulse obstruction assessment results are output. For osteoarthritis data, based on knee X-ray images, the joint space is measured using the KL grading standard and edge detection algorithm to output osteoarthritis diagnosis and grading.

3. The method according to claim 1 or 2, characterized in that, In step S3, the assessment results of each body are standardized and encoded, and then concatenated into a 128-dimensional fusion feature vector; The feature vector consists of the following features: The skin texture features are composed of color labels, temperature statistical features, and skin texture features extracted by CNN; The characteristics of muscular paralysis are composed of the rate of decrease in peak torque, the ratio of flexor to extensor torque, stability abnormality labels, and temporal characteristics of muscle strength curves. The characteristics of muscle paralysis are composed of joint angle deviation, muscle and knee elasticity labels, movement pattern disorder, and key point sequence features extracted by RNN. The characteristics of pulse obstruction are composed of blood circulation labels, pain type labels, pulse characteristics, tongue surface CNN characteristics, and pulse temporal characteristics; The characteristics of osteoarthritis are composed of KL grade, joint space width, VAS pain score and image segmentation features.

4. The method according to claim 3, characterized in that, The standardized coding includes skin dimensional coding, and the generated skin features are 24-dimensional. The features consist of: a 1-dimensional color label generated based on the skin image, a 3-dimensional temperature statistical feature calculated based on the temperature values ​​of at least three standard sites, and a 20-dimensional texture feature vector extracted from the skin image by a convolutional neural network. The standardized coding includes muscle spasm dimension coding, and the generated muscle spasm features are 28-dimensional, which consist of: a 1-dimensional peak torque reduction rate, a 1-dimensional flexion-extension torque ratio, a 1-dimensional stability anomaly label calculated based on isokinetic muscle strength test data, and a 25-dimensional temporal feature vector extracted from the muscle strength-time curve by a long short-term memory network. The standardized coding includes muscle and tendon dimension coding, and the generated muscle and tendon features are 24-dimensional. The features consist of: a 1-dimensional joint angle deviation value calculated based on the coordinate sequence of key joint points, a 1-dimensional muscle and knee elasticity label, a 1-dimensional motion pattern disorder degree, and a 21-dimensional motion temporal feature vector extracted from the coordinate sequence by a recurrent neural network. The standardized coding includes pulse obstruction dimension coding, and the generated pulse obstruction features are 24-dimensional, which consist of: a 1-dimensional blood vessel operation status label, a 1-dimensional pain type label, a 3-dimensional pulse feature value, a 6-dimensional tongue surface feature vector extracted from the tongue surface image by a convolutional neural network, and a 13-dimensional pulse time sequence feature vector extracted from the pulse time sequence data by a recurrent neural network. The standardized coding includes bone pain dimension coding, and the generated bone pain features are 28-dimensional, which consists of: 1-dimensional KL grade value, 1-dimensional joint space width measurement value, 1-dimensional VAS pain score value, and 25-dimensional image structure feature vector extracted from knee X-ray images by the image segmentation model.

5. The method according to claim 1 or 2, characterized in that, The comprehensive classification using the aforementioned classification model is specifically as follows: the fused feature vector is input into a fully connected neural network classification model; wherein, the classification model has an input layer, a first hidden layer, a second hidden layer, and an output layer connected in sequence; the number of neurons in the input layer is the same as the dimension of the fused feature vector, the first hidden layer has 64 neurons and uses the ReLU activation function, the second hidden layer has 32 neurons and uses the ReLU activation function, and the output layer has 5 neurons and uses the Sigmoid activation function, respectively outputting the existence probability values ​​corresponding to the five dimensions of physical pain: skin, muscle, tendon, vein, and bone.

6. The method according to claim 3, characterized in that, The method also includes a step of optimizing scheme generation based on association rule mining: using the Apriori algorithm to mine the association between the combination of physical pain features and traditional Chinese medicine treatment methods in historical data; storing the strong association rules that meet the conditions into the clinical rule base for matching and recommendation when generating schemes.

7. The method according to claim 3, characterized in that, The determination of the five types of physical ailments specifically includes: comparing the probability values ​​of each physical ailment output by the classification model with a preset determination threshold; if the probability value of a certain physical ailment is greater than or equal to the determination threshold, then the physical ailment is determined to exist; and combining all the physical ailment types that are determined to exist to obtain the determination result of the composite type.

8. The method according to claim 1, characterized in that, The automatic generation of the personalized TCM diagnosis and treatment plan specifically includes a rule matching step: taking the composite type judgment result and the information on the grade of each type of knee arthralgia as input, and calling a clinical-level TCM treatment method association library for matching; wherein, the association library is constructed in the following way: using TCM clinical guidelines as initial rules, and performing association verification and optimization based on the actual treatment data and efficacy feedback of historical knee arthralgia clinical cases, forming mapping rules between the combination of arthralgia characteristics, grade, TCM treatment method, and intervention intensity.

9. The method according to claim 8, characterized in that, The personalized treatment and rehabilitation plan includes the following structured output: A standardized diagnostic and identification report shall include at least the following fields: diagnostic results for the five types of arthralgia, classification, core features, and diagnostic basis based on the multi-source heterogeneous data; Personalized treatment sub-plans include at least: traditional Chinese medicine treatments with intervention intensity appropriate to the type and grade of physical pain, as well as one or more combinations of specific Chinese herbal prescriptions, acupuncture points, and physical therapy methods; A rehabilitation follow-up plan should include at least the following: standardized rehabilitation training movements, a pre-set follow-up examination cycle, and quantified indicators for evaluating the effectiveness of the treatment.

10. A system for generating auxiliary TCM diagnosis and treatment plans for knee arthritis based on the five-body hierarchy, characterized in that, include: The data acquisition module is used to collect standardized data for the five dimensions of physical pain: skin, flesh, tendons, blood vessels, and bones. A single-dimensional assessment module, connected to the data acquisition module, is used to extract and assess features of each physical pain dimension based on the acquired corresponding data and using an appropriate AI model or algorithm, and output the diagnostic results and grading of each physical pain. The fusion classification module, connected to the single-dimensional assessment module, is used to numerically encode the assessment results of each physical ailment and concatenate them to form a fusion feature vector. The fusion feature vector is then input into a trained fully connected neural network classification model to obtain the probability of the presence or absence of the five physical ailments. Based on the probability of the probability of the probability, the patient's five physical ailment complex type is determined. The treatment plan generation module, connected to the fusion classification module, is used to generate a personalized treatment plan that includes traditional Chinese medicine treatment methods, specific measures, and rehabilitation follow-up plans based on the diagnostic results of the five types of paralysis and the grade of each paralysis, combined with the patient's basic information, through rule matching.

Images