Association analysis system and method for traditional Chinese medicine constitution classification, eruptive disease feature and clinical curative effect

Abstract

The invention relates to the technical field of traditional Chinese medicine intelligent diagnosis, and discloses a traditional Chinese medicine constitution classification and eruptive therapy feature and clinical curative effect correlation analysis system and method, and the system comprises a multi-modal data collection module, an eruptive therapy feature quantitative processing module, a constitution-eruptive therapy-curative effect correlation modeling module and a result output module. The multi-modal data acquisition module is used for acquiring data such as images (eruption images), signs and the like of people with different physiques after scraping therapy in a standardized manner; the eruptive image feature quantitative processing module is used for converting the eruptive image into quantitative features; the constitution-eruptive therapy-curative effect association modeling module is used for establishing an association model of eruptive therapy features, constitution and curative effects and realizing prediction; color (12-dimensional), form (8-dimensional) and texture (10-dimensional) features are fused, Lab color space, Gabor wavelet and other technologies are applied to acute therapy image analysis for the first time, the breakthrough of acute therapy image features from'qualitative 'to'quantitative' is achieved, the feature dimension and description precision are far better than those of the prior art, and 3-5-dimensional color features are mostly adopted in the prior art.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic technology in traditional Chinese medicine, specifically to a system and method for analyzing the correlation between traditional Chinese medicine constitution classification, symptoms of sha (a type of skin disease), and clinical efficacy. Background Technology

[0002] Gua sha, a traditional Chinese medicine (TCM) health preservation technique, can regulate the body's constitution and treat ailments. It is simple, convenient, and effective. "Sha phenomena" refer to the various reactions that occur after gua sha stimulation of the skin, causing subcutaneous capillaries to dilate, deform, twist, and rupture, resulting in blood leakage and changes in skin color and shape. These changes include discoloration and morphological variations such as spots, patches, and streaks. TCM considers sha phenomena as an external sign of the human body. By observing their color, shape, and other characteristics, one can infer the circulation of qi and blood in the meridians and the functional state of corresponding organs. Sha phenomena can, to a certain extent, serve as a reference for understanding changes in the body's constitution, diagnosing and treating diseases, observing therapeutic effects, and predicting prognosis. Different constitutions produce different sha phenomena during gua sha. For example, people with blood stasis constitution develop sha rapidly, often appearing as purple or bluish-black sha spots, experiencing significant stinging sensations during gua sha, and frequently exhibiting positive reactions such as nodules. Gua sha has both therapeutic and diagnostic significance. In-depth research on the sha patterns formed after gua sha can reveal the intrinsic relationship between sha patterns and constitution, disease, and syndrome, which has important clinical significance.

[0003] However, current observation and research on petechiae (sha symptoms) mainly suffer from three core problems: First, the judgment of petechiae relies on doctors' visual observation, and descriptions such as "dark red" and "patchy" lack quantitative standards. The Kappa coefficient of consistency among different doctors is only 0.42-0.58, indicating strong subjectivity. Second, the correlation between constitution, petechiae, and therapeutic effects remains at the empirical level, lacking a quantitative correspondence between "constitutional parameters - petechiae characteristics - therapeutic indicators." Third, therapeutic effect assessment is mostly based on patients' subjective feelings, lacking support from objective indicators such as inflammatory factors and hemodynamics, and cannot predict the trend of therapeutic effect before and after treatment. Therefore, there is an urgent need for a petechiae analysis system and method that standardizes data collection, quantifies petechiae characteristics, intelligently models correlations, and objectively assesses therapeutic effects.

[0004] The "sha" (petechiae) pattern, the most obvious and intuitive visual feature formed after scraping (gua sha), includes characteristics such as color, shape, and texture. These characteristics vary and exhibit certain patterns of change among individuals with different constitutions. This study collects images of sha patterns before and after scraping treatment from individuals with different constitutions. Using computer image analysis and processing techniques, the color space distribution, morphological parameters, and texture features are extracted. Combined with TCM constitution index and clinical efficacy evaluation data, nonlinear modeling of multidimensional variables is performed using neural networks and other intelligent computer algorithms. This effectively identifies the correspondence between sha pattern characteristics and constitution classification, and predicts the trend of intervention efficacy, further validating the feasibility and scientific validity of sha patterns as a visual diagnostic basis in TCM. This provides a reference for achieving objective diagnosis and efficacy evaluation of sha patterns, and offers new research ideas and directions for the modernization of scraping techniques. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a system and method for analyzing the correlation between TCM constitution classification, sha (petechiae) characteristics, and clinical efficacy. It solves the problem that existing technologies often use the RGB color space to extract color features from sha images, neglecting multimodal and multidimensional quantitative features of morphology and texture. This invention extracts 30-dimensional quantitative features of sha, including color, morphology, and texture, and constructs a constitution-sha-efficacy correlation model. This enables objective diagnosis of constitution classification and efficacy evaluation based on sha characteristics, with high model accuracy.

[0006] (II) Technical Solution To achieve the above objectives, this invention provides the following technical solution: a system for analyzing the correlation between TCM constitution classification, sha (petechiae) characteristics, and clinical efficacy, including a multimodal data acquisition module, a sha characteristic quantification processing module, a constitution-sha-efficacy correlation modeling module, and a result output module; the multimodal data acquisition module is used to collect standardized sha and physical sign data from people with different constitutions; the sha characteristic quantification processing module is used to convert sha images into quantitative features; the constitution-sha-efficacy correlation modeling module is used to establish a correlation model between sha characteristics and constitution and efficacy and to make predictions; the result output module is used to generate visual reports and connect to clinical application scenarios.

[0007] Preferably, the multimodal data acquisition module includes hardware configuration and software support. The hardware configuration includes an MS-C200 multispectral camera, a PF-100 blood perfusion volume detector, and an intelligent scraping device. The MS-C200 multispectral camera can simultaneously acquire RGB, HSV, and 700-900nm near-infrared images. The intelligent scraping device has built-in pressure / angular velocity sensors with errors ≤0.05N and 0.1rad / s, respectively. The software support is developed based on the Windows 10 system and Python, and has functions such as subject information entry, multi-device synchronous control, data preview, and classification storage.

[0008] Preferably, the quantization processing module for the sha image features adopts a three-level process of "preprocessing-segmentation-extraction". It is developed based on Python 3.9 and integrates the OpenCV 4.5 toolkit. It processes the image by Gaussian filtering for noise reduction and histogram equalization, calls the trained U-Net model to segment the sha image region with a segmentation accuracy of 92.3%, and finally extracts 30-dimensional sha image quantization features including 12-dimensional color, 8-dimensional shape, and 10-dimensional texture.

[0009] Preferably, the constitution-symptom-treatment correlation modeling module is constructed based on a dual-branch attention neural network using TensorFlow 2.8. The constitution branch processes 16-dimensional constitution features through 3 fully connected layers, while the symptom branch processes 30-dimensional symptom features through 4 convolutional layers and 2 pooling layers. After the key correlation features are enhanced by a 4-head attention fusion layer, the dual-task output layer simultaneously outputs the constitution classification probability and treatment score. The model achieves a classification accuracy of 88.5% and a treatment prediction error of ≤3.2%.

[0010] Preferably, the result output module is based on ECharts 5.3 to develop visualization components, which can generate a pie chart of body constitution matching degree, a heat map of sha characteristics and a line chart of efficacy prediction. It provides two report templates: a professional version with 23 indicators and a popular version with 8 indicators, and supports PDF / Excel export and HL7 medical data interface docking.

[0011] The system and method for analyzing the correlation between TCM constitution classification, Sha (petechiae) characteristics, and clinical efficacy include the following steps: S1: Preparation stage, screening subjects aged 18-65 years and excluding confounding factors, and confirming constitution classification through the "TCM Constitution Assessment Scale + Dual Doctor Diagnosis"; Step 2: Data collection stage, collecting constitution, Sha (petechiae), signs, and efficacy data simultaneously at 5 time points, with 5 treatments performed at 5-day intervals; Step 3: Feature processing stage, preprocessing and segmenting the Sha (petechiae) images to extract 30-dimensional quantitative features; Step 4: Modeling and prediction stage, inputting constitution characteristics and Sha (petechiae) characteristics into a dual-branch attention neural network to obtain results; Step 5: Output stage, generating a visual report and applying it clinically.

[0012] Preferably, in the preparation stage, the Cronbach's α of the TCM constitution assessment quality scale is ≥0.85, the test-retest reliability is ≥0.82, and the Kappa coefficient of the dual-doctor diagnosis is ≥0.78; the equipment calibration includes calibrating the camera with the X-RiteColorChecker color chart and calibrating the pressure sensor of the intelligent scraping device with standard weights.

[0013] Preferably, the efficacy data in the data collection phase includes subjective symptom improvement scores and objective detection indicators. The objective detection indicators include IL-6, TNF-α inflammatory factors, and blood perfusion. The collected data is categorized and stored in JSON / DICOM3.0 / CSV format according to the "date-physical condition-ID" path.

[0014] Preferably, in the feature processing stage, color features are extracted based on the Lab color space, morphological features include roundness and contour complexity parameters, texture features are fused with gray-level co-occurrence matrix and Gabor wavelet technology, and outliers are removed from feature data using the IQR method; in the modeling stage, the Adam optimizer and joint loss function are used, with training set accuracy ≥89% and validation set accuracy ≥85%.

[0015] Preferably, the method achieves a body constitution classification accuracy of 92% and 88% for blood stasis constitution and yin deficiency constitution, respectively, with a therapeutic effect prediction error of 0.3-0.4 points, and an IL-6 decrease rate prediction error of 2.1%±0.8% and 2.3%±1.0% for subjects with blood stasis constitution and yin deficiency constitution, respectively; the prediction result can be output within 3 seconds after inputting the body constitution and sha characteristics in the inference stage.

[0016] (III) Beneficial Effects Compared with existing technologies, this invention provides a system and method for analyzing the correlation between traditional Chinese medicine constitution classification, symptoms of sha (a type of skin disease), and clinical efficacy, which has the following beneficial effects: 1. This TCM constitution classification and Sha phenomenon characteristics and clinical efficacy correlation analysis system is the first to create a 30-dimensional Sha phenomenon quantitative feature system, which integrates color (12-dimensional), morphology (8-dimensional), and texture (10-dimensional) features. It is the first to apply Lab color space, Gabor wavelet and other technologies to Sha phenomenon analysis, realizing a breakthrough from "qualitative" to "quantitative" Sha phenomenon features. The feature dimensions and description accuracy far exceed existing technologies, which are mostly 3-5 dimensional color features.

[0017] 2. This TCM constitution classification system and method, along with its correlation analysis of symptoms and clinical efficacy, constructs a constitution-symptom-efficacy correlation model. This model enables constitution classification diagnosis and efficacy evaluation based on symptoms, with high accuracy. The accuracy rates for blood stasis constitution and yin deficiency constitution reached 92% and 88% respectively, significantly higher than the traditional experience group (64%). Compared to traditional experience, which relies on visual observation of symptom color and is easily influenced by light and doctor's experience, this invention, through 30-dimensional quantitative features combined with neural networks, captures the essential differences of "low L value and high entropy value" in blood stasis constitution and "high a value and high roundness" in yin deficiency constitution, resulting in a more objective judgment.

[0018] 3. This system and method for analyzing the correlation between TCM constitution classification, characteristics of sha (scraping) patterns, and clinical efficacy provides a methodological reference for the feasibility and scientific validity of sha patterns as a visual diagnostic tool in TCM. It also provides a scientific and objective reference for the application of sha diagnosis in TCM constitution identification and efficacy evaluation, and offers new ideas and directions for the modernization research of Gua Sha therapy, a suitable TCM health preservation and rehabilitation technique. It has good clinical application value and scientific research significance. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] A system for analyzing the correlation between Traditional Chinese Medicine constitution classification, characteristics of Sha (a type of skin ailment), and clinical efficacy, including the following: Step 1: Multimodal Data Acquisition Module, Standardized Data Input Layer. This module is fundamental to ensuring the reliability of subsequent analysis. Its core innovation lies in establishing a standardized acquisition system integrating "human-machine-environment-software." Hardware configuration includes an MS-C200 multispectral camera (simultaneously acquiring RGB, HSV, and 700-900nm near-infrared images; the near-infrared band can penetrate 2mm under the skin to capture deep petechiae), a PF-100 blood perfusion analyzer (1PU accuracy, reflecting local microcirculation), and an intelligent scraping device (built-in pressure / angular velocity sensors with errors ≤0.05N and 0.1rad / s respectively). Software support (data acquisition software) is developed based on Windows 10 and Python. Core modules include subject information entry (linked to hospital ID to prevent confusion), device control (TCP / IP protocol for multi-device synchronization, latency ≤100ms), data preview (real-time image clarity verification), and storage settings (categorized and stored according to "date-physical condition-ID" path, with unified formats of JSON / DICOM3.0 / CSV). The standardized procedure requires subjects to be placed at room temperature of 25±2℃ for 10 minutes before data collection; the camera is fixed with a standard bracket (30cm away) during shooting, and the camera is calibrated with an X-RiteColorChecker color chart before each collection (color deviation ≤2%) to ensure the consistency of 200 sample data and solve the problem of "disordered collection and large error" in traditional data collection.

[0021] Step Two: Sha Image Feature Quantification Processing Module, the Core Layer of Data Analysis. This module transforms the sha image from "visual description" to "digital features," which is the core innovation of this invention. It completes the quantitative analysis through a collaborative process of "hardware acquisition - software processing." The core logic adopts a three-level workflow of "preprocessing - segmentation - extraction," relying on high-quality data provided by the image acquisition hardware, which is then accurately analyzed by the feature processing software. The software support (feature processing software) is developed based on Python 3.9 and integrates toolkits such as OpenCV 4.5 and Scikit-image 0.19.3. Core modules include image batch processing (supporting simultaneous import of 50 images, Gaussian filtering for noise reduction, and histogram equalization to enhance contrast), sha image region segmentation (calling a trained U-Net model to output a binary mask, achieving a segmentation accuracy of 92.3%), feature extraction (calculating 30-dimensional quantized features, including 12-dimensional color, 8-dimensional morphology, and 10-dimensional texture), and feature data management (results are stored as CSV, including feature name, value, and unit; extraction time ≤ 3s / image). In the key feature system, color features are based on Lab space (which improves resistance to light interference by 40%), morphological features include geometric parameters such as roundness (the roundness of blood stasis sha image is mostly ≤0.65), and texture features integrate gray-level co-occurrence matrix and Gabor wavelet to comprehensively reflect the essential attributes of sha image.

[0022] Step 3: Constitution-Sha Sign-Therapeutic Effect Correlation Modeling Module, the Core Layer of Intelligent Analysis. This module is the core of realizing the correlation between "Constitution-Sha Sign-Therapeutic Effect," based on deep learning algorithms and supported by modeling and inference software throughout the entire process. Hardware support relies on computing terminals (CPU Intel i7-12700H, GPU NVIDIA RTX 3060) to provide computing power and ensure efficient model training and inference. Software support (modeling and inference software) is built on TensorFlow 2.8, with core modules including model training (configurable learning rate, batch size, etc., real-time display of loss curves, using Adam optimizer and joint loss function), model evaluation (outputting 5-fold cross-validation accuracy, recall, etc., training set accuracy ≥89%, validation set ≥85%), and inference prediction (supporting single / batch input of 16-dimensional constitution features + 30-dimensional Sha sign features, outputting results within 3 seconds). The core model innovatively adopts a dual-branch attention neural network. The constitution branch (3 fully connected layers) processes structured data, while the sha phenomenon branch (4 convolutional layers + 2 pooling layers) processes high-dimensional features. A 4-head attention fusion layer strengthens key related features (such as blood stasis constitution and the mean weight of the a channel ≥ 0.15). The dual-task output layer simultaneously outputs the constitution classification probability and efficacy score. The model's classification accuracy reaches 88.5%, and the efficacy prediction error is ≤ 3.2%.

[0023] Step Four: Results Output Module, Clinical Application Layer. This module realizes the "clinical translation" of intelligent analysis results, meeting the differentiated needs of doctors and patients through results output software and connecting clinical application scenarios. Software Support (Results Output Software): Based on ECharts 5.3, the visualization components are developed. Core modules include data visualization (generating pie charts of constitution matching degree, heat maps of sha (a type of skin condition), and line charts of efficacy prediction), report generation and customization (providing two templates: a professional version with 23 indicators and a popular version with 8 indicators, supporting the addition of hospital logos and doctor signatures), and data integration (supporting PDF / Excel export, equipped with an HL7 medical data interface, and capable of importing into the hospital's electronic medical record system). The core functions are: outputting professional reports for doctors with feature contribution ranking to assist in diagnostic decisions; and outputting popular content such as constitution interpretation and nursing suggestions for patients to improve compliance and achieve a closed-loop transformation from "technical analysis to clinical application".

[0024] Preparation Phase: Laying the Foundation for Standardization. The core of this phase is to address the issue of "reliable data sources." Subject inclusion strictly adheres to the principle of "sample representativeness": covering all age groups from 18 to 65 years old (each group consisting of 10-year intervals), with 25 cases in each of the 8 body constitution categories, a near 1:1 male-to-female ratio, excluding confounding factors such as skin diseases and coagulation abnormalities, ensuring the samples reflect the body constitution distribution of the general population. Body constitution classification employs a dual verification method of "scale + syndrome differentiation": the 60-item scale undergoes reliability and validity testing (Cronbach's α ≥ 0.85, test-retest reliability ≥ 0.82), and syndrome differentiation is independently completed by two physicians with associate chief physician or higher titles, with a Kappa coefficient ≥ 0.78, avoiding single-judgment errors. Equipment calibration is a crucial preliminary step: the camera is calibrated using a standard color chart, and the scraping device's pressure sensor is calibrated using standard weights (0-20N) to ensure the accuracy of the collected data.

[0025] Data Acquisition Phase: Achieving "Dynamic + Multi-Source" Data Coverage. The innovation in this phase lies in the "dynamic time dimension" and the "multi-source data dimension." In the time dimension, five scraping treatments were performed, with a five-day interval between treatments. Data on constitution, petechiae (sha symptoms), physical signs, and efficacy were collected continuously at five time points, recording the dynamic changes in petechiae characteristics and data indicators. In the data dimension, five categories of data were collected simultaneously: constitution, operation, petechiae, physical signs, and efficacy. Operational parameters were standardized using intelligent devices to avoid variations in human force. Efficacy data combined subjective scoring (symptom improvement) with objective detection (inflammatory factors, blood flow velocity). IL-6 and TNF-α, as inflammatory response markers, quantify the anti-inflammatory effect of scraping, making efficacy assessment more convincing.

[0026] Feature Processing Stage: Achieving a Breakthrough in the Quantification of Sha Sign Features. This stage is the core transformation link of this invention from "experience" to "science." The construction of the 30-dimensional feature system is based on the combination of traditional Chinese medicine theory and engineering: The color feature uses the Lab space because the description of the color of Sha signs in traditional Chinese medicine (such as "purple-black" and "bright red") is closer to human vision. The a channel (red-green difference) of the Lab space can accurately quantify the difference between "red" and "purple," and the L channel (brightness) can quantify the degree of "darkness" and "brightness." The roundness and contour complexity in the morphological features correspond to the concepts of "diffuse" and "aggregated" in traditional Chinese medicine. For example, Sha signs in blood stasis constitution are more diffuse (roundness ≤ 0.65), while those in yin deficiency constitution are more aggregated (roundness ≥ 0.7). The entropy value in the texture feature can quantify the density of Sha signs. The higher the entropy value (such as 1.85±0.12 for phlegm-dampness constitution), the denser the Sha signs and the more severe the condition. After feature extraction, the IQR method is used to remove outliers to ensure data quality and provide reliable input for modeling.

[0027] Modeling Phase: Addressing the challenge of high-dimensional correlation modeling, this phase innovates by designing a dual-branch attention network to address the heterogeneity between "constitutional characteristics (low-dimensional structured)" and "symptoms of eczema (high-dimensional unstructured)." Traditional single-branch models are prone to "suppressing structured features," while the dual-branch structure allows for in-depth extraction of both types of features. The attention fusion layer automatically identifies key correlations (such as the strong correlation between blood stasis constitution and the mean of the α channel and TNF-α) by calculating mutual information weights, enabling the model to focus on features meaningful for constitution classification and therapeutic efficacy. The joint loss function simultaneously optimizes classification (constitution) and regression (therapeutic efficacy) tasks, resolving performance bias caused by single-task training. A 5-fold cross-validation and incremental training mechanism ensure the model's adaptability to different populations. When new data accumulates to ≥20%, the model parameters are updated through transfer learning, avoiding the waste of resources from retraining.

[0028] Application Phase: Achieving Clinical Implementation and Transformation. The core of this phase is "making technology serve clinical practice." Model inference only requires inputting the 16-dimensional constitution characteristics of the person being analyzed and the 30-dimensional gua sha (scraping) characteristics, and results can be output within 3 seconds. In professional reports, the feature contribution ranking can help doctors understand "why this constitution is matched," and the efficacy prediction curve can guide subsequent interventions (e.g., if the predicted efficacy score is low, it is recommended to add moxibustion as an auxiliary treatment). Popular reports use formats such as "Constitution Mini-Lessons" and "Health Guidance Tips" to improve patient compliance. After the system is connected to the HL7 interface, it can be directly integrated into the hospital's existing diagnosis and treatment processes, providing objective diagnostic tools for disease prevention centers.

[0029] It pioneered a 30-dimensional quantitative feature system for Sha symptoms, integrating color (12-dimensional), morphology (8-dimensional), and texture (10-dimensional) features. It is the first to apply Lab color space, Gabor wavelet and other technologies to Sha symptoms analysis, achieving a breakthrough from "qualitative" to "quantitative" Sha symptoms features. The feature dimensions and description accuracy far exceed existing technologies (most existing technologies are 3-5 dimensional color features).

[0030] A method for analyzing the correlation between TCM constitution classification, characteristics of Sha (a type of skin ailment), and clinical efficacy, including the following steps: S1: Subject Screening and Constitution Labeling: A 38-year-old female subject, 162cm tall and 68kg, with no skin diseases or coagulation abnormalities, was included. She completed a 60-item body composition scale (score: Yin deficiency 42 points, other constitutions ≤25 points). A deputy chief physician of traditional Chinese medicine diagnosed her with Yin deficiency based on her red tongue with little coating and thready, hesitant pulse. Her information was entered into the database as ID TSZ-038.

[0031] S2: Gua Sha Preparation: Room temperature 25℃. After the subject rests for 10 minutes, they lie face down. After cleaning their back, gua sha oil is applied. Intelligent Gua Sha Equipment Calibration: The pressure sensor is calibrated using a 5N standard weight with an error of 0.03N; Parameter settings: Force 6N, frequency 2 times / second, duration 5 minutes.

[0032] S3: Data Acquisition: ①0h: Acquire basic vital signs (temperature 36.4℃, blood perfusion 98PU) and blank skin multispectral images; ②During Gua Sha: The device records operating parameters in real time, with no abnormal fluctuations; ③5 consecutive times at 5-day intervals: Collect Gua Sha images, record vital sign data and symptom improvement scores; ④On the 5th time, collect 3mL of venous blood to detect IL-6 (8pg / mL, a decrease of 52% compared to 0h) and TNF-α (12pg / mL, a decrease of 45% compared to 0h).

[0033] S4: Image Preprocessing and Segmentation: Select the image of the bruises, and after Gaussian filtering for noise reduction and histogram equalization, the U-Net model segments the bruise region (area 18.6 cm²). 2 Morphological opening operations are used to remove back hair interference, resulting in an effective region mask.

[0034] S5: Feature Extraction: Calculate 30-dimensional features, core feature values: L=44, a=27, b=14 (color); circularity 0.61, contour complexity 0.78 (shape); entropy 1.83, contrast 0.25 (texture), and the feature vector is normalized for later use.

[0035] S6: Data Integration and Modeling: The subject's 16-dimensional constitution characteristics (high principal component score for Yin deficiency constitution), 30-dimensional Sha (symptom) characteristics, and therapeutic effect score (8 points) were integrated into a single dataset and added to the training set. During model training, the attention weights corresponding to this sample were as follows: channel a mean weight 0.16, roundness weight 0.14, which are the top 2 features.

[0036] S7: Inference and Output: Input the subject's 6-hour feature data. Model output: Constitution matching degree 92% (blood stasis constitution), efficacy prediction score 8.1 points (0.1 points error compared to the actual score of 8). Report generation: The professional version indicates that the petechiae are dark and diffuse, suggesting blood stasis obstruction, and recommends scraping once a week, combined with Danshen and safflower tea; the simplified version states that you have a blood stasis constitution, scraping is effective, and you should pay attention to more exercise.

[0037] The method of this invention achieves an accuracy rate of 92% and 88% for blood stasis constitution and yin deficiency constitution, respectively, which is significantly higher than that of the traditional experience group (64%). Compared with the traditional experience group, which relies on visual observation of the color of the petechiae and is easily affected by light and doctor's experience, this invention captures the essential differences of "low L value and high entropy value for blood stasis constitution" and "high a value and high roundness for yin deficiency constitution" through 30-dimensional quantitative features combined with neural networks, making the judgment more objective. The efficacy prediction error of the method of this invention is only 0.3-0.4 points, which is much lower than that of the traditional experience group (1.8 points). The core lies in the integration of dynamic characteristics of the petechiae and objective indicators such as inflammatory factors. Traditional experience relies solely on the patient's "feeling of improvement" to judge, while this invention achieves accurate prediction by correlating 6-hour petechiae characteristics (such as entropy value of 1.83) with the decreasing trend of IL-6.

[0038] Example: To verify the effectiveness of the present invention, 25 subjects each of blood stasis constitution and yin deficiency constitution (two typical constitutions) were selected and a comparative experiment was conducted using the system and method of the present invention. At the same time, the traditional doctor's experience judgment was used as the control group. The evaluation indicators included: the accuracy of constitution classification, the error of efficacy prediction, and the stability of characteristics (the coefficient of variation of characteristics of the same subject at different time points).

[0039] This experiment included 25 subjects from two typical constitution types: blood stasis and yin deficiency. The method of this invention was compared with traditional physician experience judgment (control group). Evaluation was conducted across four dimensions: accuracy of constitution classification, error in predicting treatment efficacy, stability of characteristics, and error in predicting the IL-6 decrease rate. Specific results are as follows: Accuracy of constitution classification: The method of this invention achieved a classification accuracy of 92.0% for subjects with blood stasis constitution (23 out of 25 cases were correctly identified), and an accuracy of 88.0% for subjects with yin deficiency constitution (22 out of 25 cases were correctly identified). In contrast, the traditional experience group achieved an overall classification accuracy of only 64.0% for 50 subjects with both constitutions (32 out of 50 cases were correctly identified). Statistical analysis showed a highly significant difference between the two groups (P < 0.01).

[0040] The efficacy prediction error (based on efficacy scores): The efficacy prediction error of the method of this invention was 0.3±0.1 points in subjects with blood stasis constitution and 0.4±0.2 points in subjects with yin deficiency constitution; the efficacy prediction error of the traditional experience group was significantly higher, reaching 1.8±0.5 points. The difference between the two groups was highly statistically significant (P value < 0.001).

[0041] Feature stability (evaluated by the coefficient of variation of features at different time points for the same subject): The coefficient of variation of the method of this invention for the characteristics of petechiae in subjects with blood stasis constitution was 3.2%±0.5%, and the coefficient of variation for subjects with yin deficiency constitution was 3.5%±0.6%, indicating excellent stability of feature data; the traditional experience group, due to its reliance on subjective judgment, had a coefficient of variation of 18.6%±2.3% for feature description, indicating extremely poor stability. The difference between the two groups was statistically significant (P value < 0.001).

[0042] IL-6 decline rate prediction error: The method of this invention can accurately predict the decline trend of inflammatory factors, with an error of 2.1%±0.8% in subjects with phlegm-dampness constitution and 2.3%±1.0% in subjects with yin deficiency constitution; while traditional experience judgment does not have this quantitative assessment indicator and cannot provide predictive data on the IL-6 decline rate.

[0043] The hardware of this system includes a multispectral camera, an intelligent scraping device, a blood perfusion analyzer, and an enzyme-linked immunosorbent assay (ELISA) analyzer. The specifications of each device are as follows: Multispectral camera: Model MS-C200, with core parameters supporting simultaneous acquisition of RGB, HSV and 700-900nm near-infrared band images, and image resolution of 3000×4000dpi; calibration method uses X-RiteColorChecker standard color chart for color correction to ensure color deviation ≤2%; calibration frequency is once after every 50 samples are acquired.

[0044] Intelligent Gua Sha device: Model SG-2024, core parameters are pressure range 0-20N (accuracy 0.05N), angular velocity adjustment range 0-10rad / s; calibration method is to calibrate the pressure sensor with 1N, 5N, and 10N standard weights, and at the same time calibrate the angular velocity parameter with an angular velocity sensor calibrator; calibration frequency is that calibration must be completed before daily use.

[0045] Blood perfusion volume analyzer: Model PF-100, core parameters are measurement range 0-500PU, measurement accuracy 1PU; calibration method uses 100Ω and 500Ω standard resistance boxes for calibration to ensure measurement error ≤2PU; calibration frequency is once a week.

[0046] Enzyme-Linked Immunosorbent Assay (ELISA) instrument: Model ELISA-800, core parameters are: detection limit 0.1 pg / mL, detection wavelength 450 nm; calibration method is to plot a standard curve by serial dilution of standards, requiring the standard curve R... 2 ≥0.99; calibration must be performed before each batch of tests.

[0047] Calibration operations must be performed by trained technicians, and calibration records must be archived for at least 3 years to ensure data traceability.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for analyzing the correlation between TCM constitution classification, symptoms of sha (a type of skin disease), and clinical efficacy, characterized by: It includes a multimodal data acquisition module, a quantification processing module for sha (petechiae) features, a constitution-sha-therapeutic effect correlation modeling module, and a results output module. The multimodal data acquisition module is used to collect sha and physical signs data of people with different constitutions in a standardized manner. The quantification processing module for sha features is used to convert sha images into quantitative features. The constitution-sha-therapeutic effect correlation modeling module is used to establish a correlation model between sha features and constitution and therapeutic effect and to achieve prediction. The results output module is used to generate visual reports and connect to clinical application scenarios.

2. The system for analyzing the correlation between TCM constitution classification, symptoms of sha (a type of skin disease), and clinical efficacy as described in claim 1, is characterized in that: The multimodal data acquisition module includes hardware configuration and software support. The hardware configuration includes an MS-C200 multispectral camera, a PF-100 blood perfusion volume detector, and an intelligent scraping device. The MS-C200 multispectral camera can simultaneously acquire RGB, HSV, and 700-900nm near-infrared images. The intelligent scraping device has built-in pressure / angular velocity sensors with errors ≤0.05N and 0.1rad / s, respectively. The software support is developed based on Windows 10 system and Python, and has the functions of subject information input, multi-device synchronous control, data preview, and classification storage.

3. The system for analyzing the correlation between TCM constitution classification, symptoms of rashes, and clinical efficacy as described in claim 1, characterized in that: The quantization processing module for the symptoms of sha (a type of skin condition) adopts a three-level process of "preprocessing-segmentation-extraction". It is developed based on Python 3.9 and integrates the OpenCV 4.5 toolkit. It processes the image by Gaussian filtering for noise reduction and histogram equalization, and calls the trained U-Net model to segment the sha region with a segmentation accuracy of 92.3%. Finally, it extracts 30-dimensional quantization features of sha, including 12-dimensional color, 8-dimensional shape, and 10-dimensional texture.

4. The system for analyzing the correlation between TCM constitution classification, symptoms of rashes, and clinical efficacy as described in claim 1, characterized in that: The constitution-symptom-treatment correlation modeling module is based on a dual-branch attention neural network constructed using TensorFlow 2.

8. The constitution branch processes 16-dimensional constitution features through 3 fully connected layers, while the symptom branch processes 30-dimensional symptom features through 4 convolutional layers and 2 pooling layers. After the key correlation features are enhanced by a 4-head attention fusion layer, the dual-task output layer simultaneously outputs the constitution classification probability and treatment score. The model achieves a classification accuracy of 88.5% and a treatment prediction error of ≤3.2%.

5. The system for analyzing the correlation between TCM constitution classification, symptoms of sha (a type of skin disease), and clinical efficacy as described in claim 1, is characterized in that: The results output module is based on ECharts 5.3 and has developed visualization components that can generate pie charts of body constitution matching degree, heat maps of sha characteristics, and line charts of efficacy prediction. It provides two report templates: a professional version with 23 indicators and a popular version with 8 indicators. It supports PDF / Excel export and HL7 medical data interface docking.

6. A system and method for analyzing the correlation between TCM constitution classification, symptoms of sha (a type of skin disease), and clinical efficacy, characterized by: Includes the following steps: S1: Preparation stage: Screening subjects aged 18-65 and excluding confounding factors, and confirming the constitution classification through the "Traditional Chinese Medicine Constitution Assessment Scale + double doctor diagnosis"; S2: Data collection stage: Five scraping treatments were performed, with one treatment every five days, and data on constitution, sha signs, physical signs and efficacy were collected simultaneously at five time points. S3: Feature processing stage, the image of sha (sha) is preprocessed and segmented to extract 30-dimensional quantized features; S4: In the modeling and prediction stage, the physical characteristics and symptoms of sha (a type of skin ailment) are input into a dual-branch attention neural network to obtain the results; S5: Output stage, generating visual reports and applying them clinically.

7. The method for correlation analysis of TCM constitution classification, sha (scraping syndrome) characteristics, and clinical efficacy as described in claim 6, characterized in that: During the preparation phase, the Cronbach's α of the Traditional Chinese Medicine Constitution Assessment Scale is ≥0.85, the test-retest reliability is ≥0.82, and the Kappa coefficient of the dual-doctor diagnosis is ≥0.

78. Equipment calibration includes calibrating the camera with the X-RiteColorChecker color chart and calibrating the pressure sensor of the intelligent scraping device with standard weights.

8. The method for correlation analysis of TCM constitution classification, sha (scraping syndrome) characteristics, and clinical efficacy as described in claim 6, characterized in that: The efficacy data in the data collection phase includes subjective symptom improvement scores and objective detection indicators. The objective detection indicators include IL-6, TNF-α inflammatory factors, and blood perfusion. The collected data is categorized and stored in JSON / DICOM3.0 / CSV format according to the "date-physical condition-ID" path.

9. The method for correlation analysis of TCM constitution classification, sha (scraping syndrome) characteristics, and clinical efficacy as described in claim 6, characterized in that: In the feature processing stage, color features are extracted based on the Lab color space, morphological features include roundness and contour complexity parameters, texture features are fused with gray-level co-occurrence matrix and Gabor wavelet technology, and outliers are removed from feature data using the IQR method; in the modeling stage, the Adam optimizer and joint loss function are used, with training set accuracy ≥89% and validation set accuracy ≥85%.

10. The method for correlation analysis of TCM constitution classification, sha (scraping syndrome) characteristics, and clinical efficacy according to claim 6, characterized in that: The method achieves an accuracy rate of 92% and 88% in classifying blood stasis constitution and yin deficiency constitution, respectively. The efficacy prediction error is 0.3-0.4 points, and the prediction error of IL-6 decrease rate is 2.1%±0.8% and 2.3%±1.0% for subjects with blood stasis constitution and yin deficiency constitution, respectively. The prediction result can be output within 3 seconds after inputting the constitution and sha characteristics in the inference stage.