Disease acupoint facies analysis system based on body surface gridding coordinate system

Abstract

The application relates to the field of traditional Chinese medicine diagnosis and treatment, and discloses a disease acupoint rash image analysis system based on a body surface gridding coordinate system. The system comprises a system data acquisition module, a body surface gridding coordinate system construction module, a rash image feature extraction and analysis module, a dynamic pathological correlation analysis module and a dynamic risk early warning module. The back body surface image, physiological signals, clinical indexes and environmental data of a patient are collected. Then, the image data is used to calibrate anatomical mark points, and grid unit codes are stored by division. The gridded body surface data is subjected to multi-dimensional color space conversion and cross-modal feature fusion to obtain quantitative rash image feature data. The rash image dynamic deviation feature set is obtained by matching a preset database according to the rash image feature data, physiological signals, clinical index data and current environmental data, the network is constructed to predict disease conversion probability and analyze acupoint correlation. Finally, the visual analysis data is obtained, a heat map and ranking are generated, and early warning information is output according to the comparison result, thereby providing strong support for traditional Chinese medicine diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of traditional Chinese medicine diagnosis and treatment technology, and more specifically to a disease acupoint sha image analysis system based on a body surface gridded coordinate system. Background Technology

[0002] Traditional Chinese medicine (TCM) analysis of skin lesions (sha) is an important diagnostic method. It mainly involves physicians visually observing the skin color, shape, and distribution characteristics of specific areas on the patient's body surface, and inferring the functional state and pathological changes of the internal organs based on TCM theory. With the development of modern medical technology, intelligent diagnostic technology combining computer vision, medical image processing, and deep learning algorithms is gradually being applied to the field of medical analysis. The aim is to improve the accuracy and scientific nature of diagnosis by objectively quantifying and recognizing biomedical signals and image data through digital means.

[0003] However, the existing technology still has the following drawbacks:

[0004] Firstly, the interpretation of traditional Chinese medicine's symptoms of sha (petechiae) relies on the physician's subjective experience and lacks standardized and quantitative assessment methods, resulting in poor consistency in diagnostic results.

[0005] Secondly, existing technologies lack the ability to dynamically and cross-modally correlate the characteristics of petechiae with disease evolution. The correlation between the shape, location, intensity, and other characteristics of petechiae and disease evolution cannot be dynamically tracked, making it difficult to achieve objective monitoring of pathological processes.

[0006] Third, the spatial positioning of acupoints and petechiae in existing technologies is not uniform, no individualized baseline of petechiae has been established, and a comparable database across patients cannot be established, making it impossible to achieve truly personalized health assessment and dynamic risk warning. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a disease acupoint sha image analysis system based on a body surface gridded coordinate system to solve the problems existing in the background art.

[0008] This invention provides the following technical solution: a disease acupoint sha image analysis system based on a body surface gridded coordinate system, comprising:

[0009] Data acquisition module: used to collect the patient's back surface imaging data, physiological signals, clinical indicator data, and current environmental data;

[0010] Body surface gridded coordinate system construction module: Based on the collected body surface image data, the anatomical landmarks on the body surface are marked, a two-dimensional rectangular coordinate system is established, and the body surface is divided into grid units for encoding and storage to obtain gridded body surface data;

[0011] The Sha (petechiae) feature extraction and analysis module is used to perform multidimensional color space conversion on gridded body surface data, extract multidimensional radiomics features, and combine them with collected physiological signals, clinical indicator data and current environmental data to perform cross-modal feature fusion to obtain quantitative Sha feature data.

[0012] Dynamic pathological correlation analysis module: Matches the characteristic data of Sha phenomenon, physiological signals, clinical indicator data and current environmental data with the preset individualized Sha phenomenon feature library to obtain the dynamic deviation feature set of Sha phenomenon. Based on the dynamic deviation feature set of Sha phenomenon, a spatiotemporal dynamic network is constructed to predict the probability of disease transformation and analyze the correlation between acupoints.

[0013] Dynamic risk warning module: It is used to receive the disease conversion probability prediction results and the correlation analysis data between acupoints, visualize and analyze the disease conversion probability prediction results and correlation analysis data, generate a decision-making basis visualization heat map and feature contribution ranking of the spatiotemporal dynamic network, and output dynamic risk warning information based on the comparison results of disease conversion probability and preset risk threshold.

[0014] Preferably, the data acquisition module acquires the patient's back surface image data through an imaging device, and simultaneously acquires the patient's physiological signals, clinical indicator data, and current environmental data; and performs time stamp synchronization and normalization processing on the back surface image data, physiological signals, clinical indicator data, and current environmental data during the acquisition process.

[0015] Preferably, the body surface gridded coordinate system construction module preprocesses the collected back body surface image data, uses the scale-invariant feature transformation algorithm to detect key points and perform matching, and achieves global alignment between different images through affine transformation or perspective transformation.

[0016] In the registered images, anatomical landmarks on the body surface are marked, and a two-dimensional rectangular coordinate system is established with the line connecting the second and third thoracic vertebrae as the Y-axis and the horizontal line of the third thoracic vertebra as the X-axis perpendicular to the Y-axis.

[0017] Based on this coordinate system, the body surface area is divided into uniform regions. The data is processed into grid cells, and each grid cell is encoded and stored to generate gridded body surface data containing anatomical location information.

[0018] Preferably, the feature extraction and analysis module extracts and analyzes the image data contained in the gridded body surface data by performing multi-dimensional color space conversion, that is, converting the RGB image to HSV, Lab and YCrCb multi-dimensional color spaces to separate hue, saturation and brightness information. For each color channel, it uses gray-level co-occurrence matrix, gray-level running length matrix and wavelet transform to extract first-order statistical features, texture features, shape features and higher-order features. At the same time, it uses LoG filter and square and logarithmic transform to enhance edge and texture details.

[0019] Principal component analysis and Pearson similarity were used to remove high-similarity features from the extracted multidimensional radiomics features for dimensionality reduction. Then, analysis of variance, LASSO and random forest algorithms were combined to screen key features related to the disease.

[0020] Finally, a cascaded convolutional neural network model was constructed. High-level features were extracted through the ResNet-50 backbone network, and EfficientNet-B4 was used to perform hierarchical prediction of lesion severity. The selected radiomics features were fused with the collected physiological signals, environmental data, and clinical indicator data across modalities to obtain quantitative bruise feature data.

[0021] Preferably, the dynamic pathological correlation analysis module matches the characteristic data of petechiae, physiological signals, clinical indicator data, and current environmental data with a preset individualized petechiae feature library to calculate the dynamic deviation feature set of petechiae. Subsequently, a Temporal Fusion Transformer spatiotemporal dynamic network is constructed, and the dynamic deviation feature set of petechiae and clinical indicator data are input at multiple time points using a sliding time window mechanism. Feature selection is performed through a gated residual network, and the temporal dependency relationship is captured using a Self-Attention mechanism to output the disease conversion probability within a future set period. At the same time, Granger causality test and structural equation model are used to analyze the temporal causal relationship and mediating effect path between the dynamic deviation feature set of petechiae and clinical indicator data to obtain the causal correlation index and path influence index. Based on the causal correlation index and path influence index, the correlation between acupoints and the influence of petechiae characteristics on disease evolution are evaluated, and the correlation analysis conclusion between acupoints is generated.

[0022] Preferably, the dynamic risk warning module is used to receive disease conversion probability, acupoint correlation analysis conclusions, causal correlation index and path influence index, and uses gradient weighted class activation mapping and SHAP value algorithm to visualize and analyze the decision basis of deep learning and time series models.

[0023] The gradient weights of the convolutional layer features are calculated to obtain the regional attention index, and a decision-making basis visualization heatmap is generated based on the regional attention index and superimposed on the original image of the bruises.

[0024] Simultaneously, the contribution of the brunt features to lesion prediction is calculated to obtain the feature attribution index, and a feature contribution ranking is generated based on the feature attribution index.

[0025] Finally, the disease conversion probability is compared with a preset risk threshold. If the disease conversion probability exceeds the preset risk threshold, it is judged as high risk and a red warning message is output. If the disease conversion probability does not exceed the preset risk threshold, it is judged as low risk and routine monitoring information is output.

[0026] The technical effects and advantages of this invention are as follows:

[0027] (1) By using the multidimensional color space conversion and image omics algorithm to transform subjective visual features into objective values ​​through the feature extraction and analysis module of Sha phenomenon, and combined with cross-modal feature fusion, the standardization and quantification of Sha phenomenon assessment are realized, effectively eliminating the bias of human interpretation and improving the consistency of diagnostic results.

[0028] (2) By constructing a spatiotemporal dynamic network through the dynamic pathological association analysis module and introducing a causal inference model, multimodal data can be deeply integrated to dynamically track the evolution trend of the symptoms of sha over time and its causal relationship with disease transformation, thereby realizing objective monitoring and accurate prediction of the pathological process.

[0029] (3) By establishing a unified anatomical coordinate system and dividing the coded grid through the body surface grid coordinate system construction module, and combining it with the individualized sha image feature library for dynamic deviation matching, the problems of inconsistent spatial positioning and lack of individualized baselines are solved, and the comparability of cross-patient data and true personalized health assessment are realized. Attached Figure Description

[0030] Figure 1 This is a system structure block diagram of the present invention. Detailed Implementation

[0031] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The disease acupoint sha image analysis system based on the body surface gridded coordinate system involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] like Figure 1 The embodiment shown provides a disease acupoint sha image analysis system based on a body surface gridded coordinate system, including:

[0033] Data acquisition module: used to collect the patient's back surface imaging data, physiological signals, clinical indicator data, and current environmental data.

[0034] In this embodiment, the data acquisition module acquires the patient's back surface image data through an imaging device, and simultaneously acquires the patient's physiological signals, clinical indicator data, and current environmental data; and performs time stamp synchronization and normalization processing on the back surface image data, physiological signals, clinical indicator data, and current environmental data during the acquisition process.

[0035] It should be specifically noted that the data acquisition module is used to acquire back surface image data of patients using high-resolution imaging equipment, simultaneously acquire physiological signals of patients using medical sensors, and input clinical indicator data from hospital information systems or clinical examination reports, as well as obtain current environmental data through environmental monitoring equipment. To ensure the accuracy of subsequent multimodal data fusion analysis, this module timestamps the back surface image data, physiological signals, clinical indicator data, and current environmental data during the acquisition process to ensure that data from different sources are aligned in the time dimension, and normalizes all acquired data to eliminate the impact of differences in data dimensions and scales on the analysis model.

[0036] Body surface gridded coordinate system construction module: Based on the collected body surface image data, anatomical landmarks on the body surface are calibrated, a two-dimensional rectangular coordinate system is established, and the body surface is divided into grid cells for encoding and storage, resulting in gridded body surface data.

[0037] In this embodiment, the body surface gridded coordinate system construction module preprocesses the collected back body surface image data, uses the scale-invariant feature transformation algorithm to detect key points and perform matching, and achieves global alignment between different images through affine transformation or perspective transformation.

[0038] In the registered images, anatomical landmarks on the body surface are marked, and a two-dimensional rectangular coordinate system is established with the line connecting the second and third thoracic vertebrae as the Y-axis and the horizontal line of the third thoracic vertebra as the X-axis perpendicular to the Y-axis.

[0039] Based on this coordinate system, the body surface area is divided into uniform regions. The data is processed into grid cells, and each grid cell is encoded and stored to generate gridded body surface data containing anatomical location information.

[0040] It should be specifically explained that the specific processing procedure of the body surface gridded coordinate system construction module is as follows: First, the acquired back surface image data is preprocessed. The Scale Invariant Feature Transform (SIFT) algorithm is used to detect and match key points between images of different patients. Then, global alignment is achieved through affine or perspective transformation to eliminate spatial errors caused by differences in patient position, height, and posture. Subsequently, the second and third thoracic vertebrae are automatically identified and labeled as anatomical landmarks in the registered images. The line connecting the second and third thoracic vertebrae is used as the Y-axis, and the horizontal line passing through the third thoracic vertebra is used as the X-axis perpendicular to the Y-axis, thus establishing a two-dimensional rectangular coordinate system for the body surface. Finally, based on this coordinate system, the body surface area is divided into uniformly spaced... The data is processed into grid cells, and each grid cell is assigned a code containing anatomical location information (such as "Feishu area-B5") for storage, generating gridded body surface data that is comparable across patients.

[0041] The Sha (petechiae) feature extraction and analysis module is used to perform multidimensional color space conversion on gridded body surface data, extract multidimensional radiomics features, and combine them with collected physiological signals, clinical indicator data and current environmental data to perform cross-modal feature fusion to obtain quantitative Sha feature data.

[0042] In this embodiment, the Sha image feature extraction and analysis module performs multi-dimensional color space conversion on the image data contained in the gridded body surface data, that is, converts the RGB image to HSV, Lab and YCrCb multi-dimensional color spaces to separate hue, saturation and brightness information. For each color channel, it uses gray-level co-occurrence matrix, gray-level running length matrix and wavelet transform to extract first-order statistical features, texture features, shape features and higher-order features. At the same time, LoG filter and square and logarithmic transform are used to enhance edge and texture details.

[0043] Principal component analysis and Pearson similarity were used to remove high-similarity features from the extracted multidimensional radiomics features for dimensionality reduction. Then, analysis of variance, LASSO and random forest algorithms were combined to screen key features related to the disease.

[0044] Finally, a cascaded convolutional neural network model was constructed. High-level features were extracted through the ResNet-50 backbone network, and EfficientNet-B4 was used to perform hierarchical prediction of lesion severity. The selected radiomics features were fused with the collected physiological signals, environmental data, and clinical indicator data across modalities to obtain quantitative bruise feature data.

[0045] It should be specifically noted that when performing multidimensional color space conversion, converting RGB images to HSV space can effectively separate hue and saturation. For example, by using the main peak position of the H channel histogram in HSV space, the descriptions of "light red," "crimson," or "dark purple" in traditional Chinese medicine (TCM) can be accurately quantified, thus deeply relating to the levels of inflammatory factors. In Lab space, the numerical deviation between the a channel (red-green axis) and the b channel (yellow-blue axis) can be used to further quantify the subtle pathological changes in the petechiae. When extracting higher-order features, wavelet transforms (such as combinations of HHH and HHL sequences) are used to decompose the segmented image into texture information of different scales and directions, which can capture the microscopic features corresponding to "ecchymosis" in TCM. When using the gray-level co-occurrence matrix to extract texture features, the entropy value is calculated to reflect the degree of disorder in the tissue, quantifying the "thickness" and heterogeneity of the petechiae. For the aforementioned dimensionality reduction and screening process, Pearson similarity is used to remove highly redundant features (such as features with a correlation coefficient greater than 0.95). Subsequently, analysis of variance (ANOVA) and LASSO regression are used to screen out key radiomics features that are significantly related to the disease, effectively solving the problem of feature dimensionality explosion and improving the robustness of subsequent models. When constructing the cascaded convolutional neural network model, the first stage, the ResNet-50 backbone network, extracts deep semantic features of the image through the residual module for pre-screening the presence of lesions. The second stage, the EfficientNet-B4 network, performs fine-grained stratification of the screened positive samples (such as mild / moderate / severe). Finally, the deep features are concatenated or weighted with non-image features such as physiological signals and environmental data to generate quantitative lesion feature data containing multi-dimensional information, providing comprehensive data support for subsequent disease prediction.

[0046] Dynamic pathological association analysis module: Matches the characteristic data of Sha phenomenon, physiological signals, clinical indicator data and current environmental data with the preset individualized Sha phenomenon feature library to obtain the dynamic deviation feature set of Sha phenomenon. Based on the dynamic deviation feature set of Sha phenomenon, a spatiotemporal dynamic network is constructed to predict the probability of disease transformation and analyze the correlation between acupoints.

[0047] In this embodiment, the dynamic pathological correlation analysis module matches the characteristic data of petechiae, physiological signals, clinical indicator data, and current environmental data with a preset individualized petechiae feature library to calculate the dynamic deviation feature set of petechiae. Subsequently, a Temporal Fusion Transformer spatiotemporal dynamic network is constructed, and a sliding time window mechanism is used to input the dynamic deviation feature set of petechiae and clinical indicator data at multiple time points. Feature selection is performed through a gated residual network, and a Self-Attention mechanism is used to capture temporal dependencies, outputting the probability of disease transformation within a set future period. At the same time, Granger causality test and structural equation model are used to analyze the temporal causal relationship and mediating effect path between the dynamic deviation feature set of petechiae and clinical indicator data, obtaining the causal correlation index and path influence index. Based on the causal correlation index and path influence index, the correlation between acupoints and the influence of petechiae characteristics on disease evolution are evaluated, generating the correlation analysis conclusion between acupoints.

[0048] It should be specifically explained that the dynamic pathological correlation analysis module processes the following steps: First, it matches the characteristic data of petechiae, physiological signals, clinical indicators, and current environmental data with a pre-defined individualized petechiae feature library to calculate a dynamic deviation feature set of petechiae. This deviation feature set includes, for example, the rate of change of hue (AH) and the fractal dimension drift trajectory (FD value). Subsequently, it constructs Temporal Fusion... The Transformer spatiotemporal dynamic network utilizes a sliding time window mechanism to input the dynamic deviation feature set of bruises and clinical indicator data at multiple time points, such as setting a window length of 3 months and a step size of 1 month, to capture the changing trends at different time scales. It employs a gated residual network for feature selection to reduce redundant information and utilizes a self-attention mechanism to capture temporal dependencies and identify key time nodes, thereby outputting the probability of disease transformation within a future set period (e.g., the next 6 months). Simultaneously, it uses Granger causality tests and structural equation modeling to analyze the temporal causal relationship and mediating effect paths between the dynamic deviation feature set of bruises and clinical indicator data. For example, it uses dynamic bruise features as independent variables and indicators of malignant transformation of pulmonary nodules (such as volume doubling time, VDT) as the basis for analysis. Using 400 days as the dependent variable, the F-statistic and P-value were used to determine whether to reject the null hypothesis. The Sobel test was used to calculate the mediating effect of inflammatory microenvironment factors, obtaining the causal association index and path influence index. Finally, based on the causal association index and path influence index, the influence of acupoint associations and petechiae characteristics on disease evolution was assessed, generating conclusions from the acupoint association analysis. Specifically, the conclusions from the acupoint association analysis refer to the quantitative assessment results of the causal relationship and interaction degree of different acupoint petechiae characteristics in the disease development process. For example, the conclusion obtained through Granger causality test may indicate that "the color change of petechiae (AH) in the Lung Shu area is statistically significantly ahead of the malignant transformation of nodules (P... The path influence index calculated by structural equation modeling showed that the direct influence coefficient of the Lung Shu area on nodule transformation was 0.6, while the indirect influence coefficient mediated by inflammatory factors was 0.4, thus clarifying the primary and secondary status of this acupoint in disease early warning and its synergistic mechanism with other acupoints.

[0049] Dynamic risk warning module: It is used to receive the disease conversion probability prediction results and the correlation analysis data between acupoints, visualize and analyze the disease conversion probability prediction results and correlation analysis data, generate a decision-making basis visualization heat map and feature contribution ranking of the spatiotemporal dynamic network, and output dynamic risk warning information based on the comparison results of disease conversion probability and preset risk threshold.

[0050] In this embodiment, the dynamic risk warning module is used to receive the disease conversion probability, the conclusion of the correlation analysis between acupoints, the causal correlation index and the path influence index, and to use gradient weighted class activation mapping and SHAP value algorithm to visualize and analyze the decision basis of deep learning and time series models.

[0051] The gradient weights of the convolutional layer features are calculated to obtain the regional attention index, and a decision-making basis visualization heatmap is generated based on the regional attention index and superimposed on the original image of the bruises.

[0052] Simultaneously, the contribution of the brunt features to lesion prediction is calculated to obtain the feature attribution index, and a feature contribution ranking is generated based on the feature attribution index.

[0053] Finally, the disease conversion probability is compared with a preset risk threshold. If the disease conversion probability exceeds the preset risk threshold, it is judged as high risk and a red warning message is output. If the disease conversion probability does not exceed the preset risk threshold, it is judged as low risk and routine monitoring information is output.

[0054] It should be specifically explained that the implementation process of the dynamic risk warning module is as follows: First, the module receives the disease conversion probability, acupoint correlation analysis conclusions, causal correlation index, and path influence index output by the dynamic pathological association analysis module. Gradient-weighted class activation mapping and SHAP value algorithms are used to visualize and analyze the decision-making basis of the deep learning and time-series models. During the heatmap generation process, a high-level convolutional layer of the deep learning model (such as the conv5_x layer of ResNet) is selected as the target layer. The gradient weights of the feature maps are calculated to obtain the regional attention index. Based on the regional attention index, a decision-making basis visualization heatmap superimposed on the original bruise image is generated to intuitively display the model's judgment. Key areas of concern identified as high-risk (such as the Lung Shu or Heart Shu areas) are identified. Simultaneously, the contribution of petechiae features to lesion prediction is calculated using Monte Carlo sampling-optimized TreeSHAP or DeepSHAP methods to obtain a feature attribution index. A feature contribution ranking is then generated based on this index (e.g., the ranking shows that "hue intensity" has the highest contribution, followed by "texture roughness," thus explaining the source of risk). Finally, the disease conversion probability is compared with a preset risk threshold. If the disease conversion probability exceeds the preset risk threshold, it is classified as high-risk and a red warning is issued; otherwise, it is classified as low-risk and routine monitoring information is issued.

[0055] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0056] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A disease acupoint sha (palpitation) image analysis system based on a body surface gridded coordinate system, characterized in that, include: Data acquisition module: used to collect the patient's back surface imaging data, physiological signals, clinical indicator data, and current environmental data; Body surface gridded coordinate system construction module: Based on the collected body surface image data, the anatomical landmarks on the body surface are marked, a two-dimensional rectangular coordinate system is established, and the body surface is divided into grid units for encoding and storage to obtain gridded body surface data; The Sha (petechiae) feature extraction and analysis module is used to perform multidimensional color space conversion on gridded body surface data, extract multidimensional radiomics features, and combine them with collected physiological signals, clinical indicator data and current environmental data to perform cross-modal feature fusion to obtain quantitative Sha feature data. Dynamic pathological correlation analysis module: Matches the characteristic data of Sha phenomenon, physiological signals, clinical indicator data and current environmental data with the preset individualized Sha phenomenon feature library to obtain the dynamic deviation feature set of Sha phenomenon. Based on the dynamic deviation feature set of Sha phenomenon, a spatiotemporal dynamic network is constructed to predict the probability of disease transformation and analyze the correlation between acupoints. Dynamic risk warning module: It is used to receive the disease conversion probability prediction results and the correlation analysis data between acupoints, visualize and analyze the disease conversion probability prediction results and correlation analysis data, generate a decision-making basis visualization heat map and feature contribution ranking of the spatiotemporal dynamic network, and output dynamic risk warning information based on the comparison results of disease conversion probability and preset risk threshold.

2. The disease acupoint sha image analysis system based on a body surface gridded coordinate system according to claim 1, characterized in that, The data acquisition module acquires back surface image data of the patient through imaging equipment, and simultaneously acquires the patient's physiological signals, clinical indicators, and current environmental data; during the acquisition process, the back surface image data, physiological signals, clinical indicators, and current environmental data are time-stamped, synchronized, and normalized.

3. The disease acupoint sha image analysis system based on a body surface gridded coordinate system according to claim 2, characterized in that, The body surface gridded coordinate system construction module preprocesses the collected back body surface image data, uses the scale-invariant feature transformation algorithm to detect key points and perform matching, and achieves global alignment between different images through affine transformation or perspective transformation. In the registered images, anatomical landmarks on the body surface are marked, and a two-dimensional rectangular coordinate system is established with the line connecting the second and third thoracic vertebrae as the Y-axis and the horizontal line of the third thoracic vertebra as the X-axis perpendicular to the Y-axis. Based on this coordinate system, the body surface area is divided into uniform regions. The data is processed into grid cells, and each grid cell is encoded and stored to generate gridded body surface data containing anatomical location information.

4. The disease acupoint sha image analysis system based on a body surface gridded coordinate system according to claim 3, characterized in that, The Sha image feature extraction and analysis module performs multi-dimensional color space conversion on the image data contained in the gridded body surface data, that is, converts the RGB image to HSV, Lab and YCrCb multi-dimensional color spaces to separate hue, saturation and brightness information. For each color channel, it uses gray-level co-occurrence matrix, gray-level running length matrix and wavelet transform to extract first-order statistical features, texture features, shape features and higher-order features. At the same time, LoG filter and square and logarithmic transform are used to enhance edge and texture details. Principal component analysis and Pearson similarity were used to remove high-similarity features from the extracted multidimensional radiomics features for dimensionality reduction. Then, analysis of variance, LASSO and random forest algorithms were combined to screen key features related to the disease. Finally, a cascaded convolutional neural network model was constructed. High-level features were extracted through the ResNet-50 backbone network, and EfficientNet-B4 was used to perform hierarchical prediction of lesion severity. The selected radiomics features were fused with the collected physiological signals, environmental data, and clinical indicator data across modalities to obtain quantitative bruise feature data.

5. The disease acupoint sha image analysis system based on a body surface gridded coordinate system according to claim 4, characterized in that, The dynamic pathological correlation analysis module matches the characteristic data of petechiae, physiological signals, clinical indicators, and current environmental data with a preset individualized petechiae feature library to calculate the dynamic deviation feature set of petechiae. Subsequently, a Temporal Fusion Transformer spatiotemporal dynamic network is constructed, and a sliding time window mechanism is used to input the dynamic deviation feature set of petechiae and clinical indicator data at multiple time points. Feature selection is performed through a gated residual network, and a Self-Attention mechanism is used to capture temporal dependencies, outputting the probability of disease transformation within a set future period. At the same time, Granger causality test and structural equation model are used to analyze the temporal causal relationship and mediating effect path between the dynamic deviation feature set of petechiae and clinical indicator data, obtaining the causal correlation index and path influence index. Based on the causal correlation index and path influence index, the correlation between acupoints and the influence of petechiae characteristics on disease evolution are evaluated, generating the correlation analysis conclusion between acupoints.

6. The disease acupoint sha image analysis system based on a body surface gridded coordinate system according to claim 5, characterized in that, The dynamic risk warning module is used to receive disease conversion probability, acupoint correlation analysis conclusions, causal correlation index and path influence index, and uses gradient weighted class activation mapping and SHAP value algorithm to visualize and analyze the decision basis of deep learning and time series models. The gradient weights of the convolutional layer features are calculated to obtain the regional attention index, and a decision-making basis visualization heatmap is generated based on the regional attention index and superimposed on the original image of the bruises. Simultaneously, the contribution of the brunt features to lesion prediction is calculated to obtain the feature attribution index, and a feature contribution ranking is generated based on the feature attribution index. Finally, the disease conversion probability is compared with a preset risk threshold. If the disease conversion probability exceeds the preset risk threshold, it is judged as high risk and a red warning message is output. If the disease conversion probability does not exceed the preset risk threshold, it is judged as low risk and routine monitoring information is output.

Images