Skin microvascular inflammation prediction method and system

Abstract

The invention relates to a skin microvascular inflammation prediction method and system, and belongs to the technical field of artificial intelligence, and the method comprises the steps: obtaining health monitoring data of a user, carrying out the preprocessing of the health monitoring data, and obtaining the preprocessed health monitoring data, the health monitoring data comprises skin image data, user health data, user current environment data and user physiological parameters; predicting the preprocessed health monitoring data based on a deep learning model to obtain a first inflammation risk score of the user; determining a first risk level of the user according to the first inflammation risk score and the dynamic threshold range; and under the condition that the triggering rule condition is met, verifying the first risk level based on the health influence parameter of the user to obtain a target risk level of the user, performing early warning notification based on the target risk level, and generating personalized suggestions based on the target risk level and the health monitoring data. The method can accurately predict the skin microvascular inflammation.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a method and system for predicting skin microvascular inflammation. Background Art

[0002] The skin microvascular network is a vital component of the human circulatory system, and its status and function directly reflect overall health. Changes in microvascular parameters such as morphology, density, and blood flow velocity are often closely associated with various inflammatory, metabolic, and circulatory diseases. Currently, monitoring of skin microvasculature relies primarily on traditional methods such as microscopy and laser Doppler velocimetry. These methods have the following major drawbacks: Traditional equipment has limited resolution and dynamic monitoring capabilities for microvasculature, making it difficult to obtain high-precision images of microvasculature and their dynamic changes in real time; data processing is complex: Manual analysis of microvascular images is not only time-consuming and labor-intensive, but also highly subjective, making it difficult to ensure consistent and accurate results; and there is a lack of comprehensive analysis: Existing systems often rely on single-source data analysis and are unable to effectively integrate multiple data sources (such as images, user health data, and environmental parameters). This results in limited predictive capabilities and makes it difficult to accurately predict inflammation risk. Therefore, how to better predict skin microvascular inflammation has become an urgent issue. Summary of the Invention

[0003] In view of the above shortcomings of the prior art, the purpose of the present invention is to provide a method and system for predicting skin microvascular inflammation. The method can accurately predict skin microvascular inflammation.

[0004] In a first aspect of the present invention, a method for predicting skin microvascular inflammation is proposed, comprising: S1, obtaining health monitoring data of a user, and preprocessing the health monitoring data to obtain the preprocessed health monitoring data, wherein the health monitoring data includes skin image data, user health data, user current environment data and user physiological parameters; S2, predicting the preprocessed health monitoring data based on a deep learning model to obtain a first inflammation risk score of the user; S3, determining a first risk level of the user based on the first inflammation risk score and a range of a dynamic threshold, wherein the first risk level includes low risk, medium risk and high risk; S4, when the trigger rule conditions are met, verifying the first risk level based on the health impact parameters of the user to obtain the target risk level of the user, and performing an early warning notification based on the target risk level and generating personalized recommendations based on the target risk level and the health monitoring data.

[0005] Furthermore, the preprocessed health monitoring data is predicted based on a deep learning model to obtain the first inflammation risk score of the user, including: obtaining a first eigenvector of the preprocessed skin image data, a second eigenvector of the user health data, and a third eigenvector of the user's current environment data and the user's physiological parameters; fusing the first eigenvector, the second eigenvector and the third eigenvector to obtain a target eigenvector; inputting the target eigenvector into the deep learning model to output the first inflammation risk score of the user.

[0006] Further, the first feature vector, the second feature vector and the third feature vector are fused to obtain a target feature vector, including: based on , calculate the weight of the first eigenvector; based on , calculate the weight of the second eigenvector; based on , calculate the weight of the third eigenvector, where represents the first eigenvector, represents the weight vector corresponding to the first eigenvector, represents the bias vector of the first eigenvector, represents the second eigenvector, represents the weight vector corresponding to the second eigenvector, represents the bias vector of the second eigenvector, represents the third eigenvector, represents the weight vector corresponding to the third eigenvector, Represents the bias vector of the third eigenvector, T represents the transpose operation of the matrix, and exp represents the activation function; based on , performing weighted summation on the weight of the first feature vector, the weight of the second feature vector, and the weight of the third feature vector to obtain the target feature vector; wherein, inputting the target feature vector into the deep learning model to output the first inflammation risk score of the user comprises: inputting the target feature vector into multiple attention heads in the deep learning model, wherein the multiple attention heads generate a query vector Q, a key vector K, and a value vector V by performing a linear transformation on the target feature vector; according to , get the attention weight of each attention head, where T represents the length of the data sequence, Represents the dimension of the key vector; according to , the attention weights of each attention head are concatenated and linearly transformed to obtain fusion information, where h represents the number of attention heads, W o Represents the output transformation matrix; based on , the first calculation of residual connection and layer normalization, where Z represents the fusion information, Represents the input data sequence; based on , calculate the feedforward network, where W1, W2, 、 To learn the parameters, Represents the characteristic matrix during the operation; based on , calculate the residual connection and layer normalization again, where, Represents the encoder output of the current layer; based on , obtaining the target features output by the multi-layer encoder; inputting the target features into the fully connected layer in the deep learning model to obtain the first inflammation risk score of the user, where F represents the fused feature vector.

[0007] Further, determining the range of the dynamic threshold includes: based on , calculate the risk offset, where Q represents the risk offset, represents the risk contribution value of each user feature, Represents the weight of each user feature, wherein the user features include user health data, current environment data and physiological parameters; the sum of the preset initial threshold and the risk offset is used as the dynamic threshold; the range of the dynamic threshold is determined based on the dynamic threshold and the upper and lower fluctuations, or based on , calculate the risk offset, where n represents the length of the input t vector, i and j represent the current cursor positions, K represents the total length of the vector, W represents the weight of all user features, and ω represents the trainable weight calculation. Represents manually specified non-trainable weights used to balance the deviations between multiple samples. represents the exponential decay factor, , represents the kth column of t, represents the j-th column of t, express The i-th column of represents the i-th column of ω; based on , calculate the dynamic threshold, Indicates the preset initial threshold, represents the sum of the thresholds of multiple target risks, , Indicates the threshold for calculating target risk 1, Indicates the threshold for calculating target risk 2, Indicates the threshold for calculating target risk 3, Represents a threshold for calculating the target risk n; determines the range of the dynamic threshold based on the dynamic threshold and the upper and lower fluctuation amounts; wherein, according to the first inflammation risk score and the range of the dynamic threshold, determines the first risk level of the user, including: when it is judged that the first inflammation risk score is lower than the lower limit of the range of the dynamic threshold, determines that the first risk level of the user is low risk; when it is judged that the first inflammation risk score is within the range of the dynamic threshold, determines that the first risk level of the user is medium risk; when it is judged that the first inflammation risk score is higher than the upper limit of the range of the dynamic threshold, determines that the first risk level of the user is high risk.

[0008] Furthermore, the trigger rule conditions include environmental impact rules, medical history impact rules, and comprehensive factor impact rules. When the trigger rule conditions are met, the first risk level is verified based on the health impact parameters of the user to obtain the target risk level of the user, including: when any one of the trigger rule conditions is met, the health impact parameters of the user are obtained; under the health impact parameters of the user, based on , calculate the adjustment coefficient for the first risk level, where represents the scaling factor, A represents the result of nonlinear activation of the original score, , S represents the original score, , a+b+c represents the total number of items, which means the number of input variables to be considered, and m represents the index variable to be summed. represents the weight coefficient of the input variable, represents an input variable; based on the adjustment coefficient of the first risk level, the first risk level is used as the target risk level, or the first risk level is increased and the increased first risk level is used as the target risk level, or the first risk level is reduced and the reduced first risk level is used as the target risk level.

[0009] Furthermore, the method further includes: obtaining a weight value of each trigger rule condition when multiple trigger rule conditions are simultaneously satisfied; obtaining a risk level under each trigger rule condition based on the weight value of each trigger rule condition; taking the sum of the risk levels under each trigger rule condition as the target risk level; or obtaining a weight value of each trigger rule condition when multiple trigger rule conditions are simultaneously satisfied; obtaining a risk level based on the weight value of each trigger rule condition; , calculate the target risk level, where h represents the number of trigger rule conditions, Indicates the triggering rule conditions The weight value of Indicates the triggering rule conditions The confidence level of the current user feature, Indicates mapping discrete levels to numerical values, i and j represent the current cursor position, Represents the interaction coefficient of multiple trigger rule conditions, Indicates the interactive confidence of multiple trigger rule conditions, Indicates the similarity between multiple trigger rule conditions.

[0010] Furthermore, the personalized suggestions include diet adjustment, exercise suggestions, environmental improvement and medical examination suggestions.

[0011] In the second aspect of the present invention, a skin microvascular inflammation prediction system is proposed, including: an acquisition module for acquiring the user's health monitoring data and preprocessing the health monitoring data to obtain the preprocessed health monitoring data, wherein the health monitoring data includes skin image data, user health data, user current environment data and user physiological parameters; a first acquisition module for predicting the preprocessed health monitoring data based on a deep learning model to obtain the user's first inflammation risk score; a determination module for determining the user's first risk level based on the first inflammation risk score and the range of the dynamic threshold, wherein the first risk level includes low risk, medium risk and high risk; a second acquisition module for verifying the first risk level based on the user's health impact parameters when the trigger rule conditions are met to obtain the user's target risk level, and performing early warning notification based on the target risk level and generating personalized suggestions based on the target risk level and the health monitoring data.

[0012] In a third aspect of the present invention, an electronic device is proposed, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any one of the methods described in the first aspect of the present invention.

[0013] According to a fourth aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any one of the methods according to the first aspect of the present invention.

[0014] The beneficial effects of the present invention are as follows: The skin microvascular inflammation prediction method and system described in the present invention obtains the user's health monitoring data and preprocesses the health monitoring data to obtain preprocessed health monitoring data, which includes skin image data, user health data, user current environment data, and user physiological parameters; predicts the preprocessed health monitoring data based on a deep learning model to obtain the user's first inflammation risk score; determines the user's first risk level based on the first inflammation risk score and the range of dynamic thresholds, and the first risk level includes low risk, medium risk, and high risk; when the trigger rule conditions are met, verifies the first risk level based on the user's health impact parameters to obtain the user's target risk level, and issues an early warning notification based on the target risk level, and generates personalized recommendations based on the target risk level and health monitoring data. This method can accurately predict skin microvascular inflammation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are only for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. Throughout the drawings, the same reference numerals represent the same components. Obviously, the drawings described below are only some of the embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings.

[0016] Figure 1 is a flow chart of a method for predicting skin microvascular inflammation according to one embodiment of the present invention; Figure 2 is a flow chart of a method for predicting skin microvascular inflammation according to a specific embodiment of the present invention; Figure 3 is a structural block diagram of a skin microvascular inflammation prediction system according to an embodiment of the present invention; Figure 4 It is a structural block diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work should fall within the scope of protection of the present invention.

[0018] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0019] In the description of the present invention, it should be noted that, unless otherwise expressly specified and limited, the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second" and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance. The terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the internal parts of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0020] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of methods and systems consistent with certain aspects of the present invention, as detailed in the appended claims.

[0021] The present invention proposes a method, system and related equipment for predicting skin microvascular inflammation. Specifically, the method, system and related equipment for predicting skin microvascular inflammation according to an embodiment of the present invention are described below with reference to the accompanying drawings.

[0022] Figure 1 This is a flow chart of a method for predicting skin microvascular inflammation according to one embodiment of the present invention. It should be noted that the method for predicting skin microvascular inflammation according to this embodiment of the present invention can be applied to the system for predicting skin microvascular inflammation according to this embodiment of the present invention. This system for predicting skin microvascular inflammation can be configured on an electronic device or on a server. This embodiment of the present application is not limited to this.

[0023] like Figure 1 As shown, the prediction method of skin microvascular inflammation includes: S110, obtaining the user's health monitoring data and preprocessing the health monitoring data to obtain the preprocessed health monitoring data, where the health monitoring data includes skin image data, user health data, user current environment data, and user physiological parameters.

[0024] In embodiments of the present invention, skin image data can be acquired using specialized skin image acquisition equipment (e.g., a high-resolution skin camera or smart device). Furthermore, OCT technology can be used to acquire three-dimensional structural images of skin microvessels, i.e., skin image data.

[0025] In embodiments of the present invention, personal health information, including past medical history, family medical history, lifestyle habits (e.g., diet, exercise), medication use, etc., can be input through a mobile application or webpage to obtain user health data. User health data can also be obtained through an interface with an electronic health record system.

[0026] In an embodiment of the present invention, the user's current environmental data can be collected in real time through environmental sensors or user devices (such as smart bracelets).

[0027] In an embodiment of the present invention, a smart bracelet or other health monitoring device can be used to automatically collect user physiological parameters, including but not limited to heart rate, blood oxygen saturation, etc.

[0028] In an embodiment of the present invention, when skin image data, user health data, user current environment data and user physiological parameters are obtained based on various devices, the obtained skin image data, user health data, user current environment data and user physiological parameters can be synchronized and transmitted through Wi-Fi, Bluetooth, 4G / 5G networks, and the user's health monitoring data can be obtained based on the terminal device.

[0029] In an embodiment of the present invention, when the user's health monitoring data is obtained, the user's health monitoring data may be pre-processed. Specifically: When skin image data is acquired, random noise can be removed and image quality improved by applying image denoising algorithms such as median filtering and Gaussian filtering. Histogram equalization and contrast stretching can be used to enhance image contrast and edge clarity, highlighting microvascular structures. Deep learning-based image segmentation algorithms (such as U-Net and Mask R-CNN) can be used to accurately segment the skin microvascular region and extract its morphological features.

[0030] Among them, when user health data is obtained, missing values and abnormal values can be processed, missing data can be filled through interpolation or other completion methods, and abnormal data can be eliminated or corrected; data of different dimensions can be converted into a unified standard, such as normalization.

[0031] Among them, when the user's current environmental data and user physiological parameters are obtained, the timestamps of different data sources can be aligned to ensure the temporal consistency of the data, which is convenient for subsequent time series analysis; principal component analysis (PCA), t-SNE and other technologies are used to reduce the dimensionality of high-dimensional data, extract key features, and reduce redundant information.

[0032] For high-dimensional data, key insights are extracted from user health data, current environmental data, and physiological parameters, and then dimensionality reduction is performed. High-dimensional data can be understood as data containing multiple variables or features. For example, environmental parameters may include temperature, humidity, and light intensity, while physiological parameters may include heart rate and blood oxygen saturation. In their original form, this data may have high dimensionality, but some dimensions may be redundant or have little contribution to the task objectives.

[0033] Among them, the significance of dimensionality reduction is to reduce redundant features, retain core information, reduce computational complexity, and improve model performance.

[0034] Among them, the implementation methods of dimensionality reduction include using principal component analysis (PCA) to project high-dimensional data into low-dimensional space through linear transformation to retain the main variation information; using nonlinear dimensionality reduction (such as t-SNE) to capture the nonlinear relationship of data, which is suitable for data with complex distributions; using statistical methods or feature importance (based on feature importance scoring of random forest or XGBoost) to screen out the features that contribute most to the task.

[0035] S120: Predict the preprocessed health monitoring data based on the deep learning model to obtain the user's first inflammation risk score.

[0036] In an embodiment of the present invention, a first eigenvector of preprocessed skin image data, a second eigenvector of user health data, and a third eigenvector of the user's current environment data and physiological parameters are obtained; the first, second, and third eigenvectors are fused to obtain a target eigenvector; the target eigenvector is input into a deep learning model to output the user's first inflammation risk score. For specific implementation methods, please refer to the subsequent embodiments.

[0037] In an embodiment of the present invention, deep learning model training and optimization are implemented as follows: 1) Training Data Preparation: Data Annotation: Collect and annotate a large number of multimodal data samples, including labels for normal microvascular states and inflammatory states. Data Augmentation: Increase the diversity of training data through data augmentation techniques (such as rotation, scaling, and flipping) to improve the model's generalization capabilities. Data Partitioning: Divide the dataset into training, validation, and test sets to ensure effective model training and accurate evaluation.

[0038] 2) Loss Function Design: For classification tasks, use the cross-entropy loss function to optimize the model's performance on classification tasks. For regression tasks, use the mean squared error (MSE) loss function to optimize the model's performance on regression tasks. Multi-task Learning: If the system performs both classification and regression tasks, a multi-task loss function can be designed to comprehensively optimize the model's performance across multiple aspects.

[0039] 3) Optimization Algorithm: Adam Optimizer: Uses the Adam or AdamW optimization algorithms to iteratively update model parameters, improving model convergence speed and prediction performance. Learning Rate Scheduling: Dynamically adjusts the learning rate based on training progress, such as using cosine annealing or learning rate decay strategies, to improve training effectiveness.

[0040] 4) Regularization and Overfitting Prevention: Dropout: Dropout layers are added to fully connected layers and Transformer layers to randomly drop some neurons to prevent model overfitting. L2 regularization: L2 regularization is added to the loss function to limit the size of model weights and improve model generalization. Early stopping: When performance on the validation set stops improving, training is stopped early to prevent the model from overfitting on the training set.

[0041] 5) Model Evaluation and Validation: Cross-Validation: K-fold cross-validation is used to evaluate the model's performance on different datasets to ensure its stability and reliability. Performance Metrics: Classification: Accuracy, recall, F1 score, and area under the receiver operating characteristic (ROC) curve (AUC) are used to comprehensively evaluate the model's classification performance. Regression: Mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) are used to evaluate the model's regression performance. Confusion Matrix Analysis: The confusion matrix is used to analyze the model's predictions for different categories and identify model biases and deficiencies. Hyperparameter Optimization: Optimize the model's hyperparameters and improve overall performance through methods such as grid search, random search, or Bayesian optimization.

[0042] S130: Determine a first risk level of the user based on the first inflammation risk score and the range of the dynamic threshold, where the first risk level includes low risk, medium risk, and high risk.

[0043] In embodiments of the present invention, once a dynamic threshold is calculated, a range of the dynamic threshold can be determined based on the dynamic threshold and the fluctuation. Furthermore, by comparing the first inflammation risk score with the dynamic threshold range, the user's first risk level can be determined. For specific implementations, please refer to the subsequent embodiments.

[0044] S140, when the trigger rule conditions are met, the first risk level is verified based on the user's health impact parameters to obtain the user's target risk level, and an early warning notification is issued based on the target risk level and personalized suggestions are generated based on the target risk level and health monitoring data.

[0045] In an embodiment of the present invention, the trigger rule conditions include environmental influence rules, medical history influence rules, and comprehensive factor influence rules.

[0046] In an embodiment of the present invention, the first risk level may be verified based on one or more of the trigger rule conditions. Specific implementation methods may refer to subsequent embodiments.

[0047] According to an embodiment of the present invention, a method for predicting skin microvascular inflammation obtains and preprocesses a user's health monitoring data to obtain preprocessed health monitoring data. The health monitoring data includes skin image data, user health data, user current environment data, and user physiological parameters. A deep learning model is used to predict the preprocessed health monitoring data to obtain a first inflammation risk score for the user. Based on the first inflammation risk score and a dynamic threshold range, the user's first risk level is determined, which includes low, medium, and high risk. If trigger rule conditions are met, the first risk level is verified based on the user's health impact parameters to obtain a target risk level for the user. Warning notifications are issued based on the target risk level, and personalized recommendations are generated based on the target risk level and health monitoring data. By acquiring health monitoring data, including skin image data, user health data, user current environment data, and user physiological parameters, this method comprehensively considers multiple factors that may affect skin microvascular inflammation. This multi-dimensional data fusion provides a more comprehensive picture of the user's health status and improves prediction accuracy compared to prediction methods that rely solely on a single type of data (such as skin images). Using a deep learning model to predict the preprocessed data automatically learns complex features and patterns within the data. Deep learning models possess powerful representational capabilities when processing large amounts of complex data, enabling them to uncover the potential nonlinear relationship between health monitoring data and the risk of skin microvascular inflammation. A user's primary risk level is determined based on the primary inflammation risk score combined with a dynamic threshold range, a method that adapts well to individual differences and environmental variations. When triggering rule conditions are met, the primary risk level is validated against the user's health impact parameters to determine a target risk level. This step further optimizes the risk assessment process by fully considering other health-related factors that influence risk levels, such as recent use of new products that may trigger inflammation and the presence of new medical conditions. This avoids potential bias associated with relying solely on a single assessment method, resulting in more accurate risk level determination. Personalized recommendations are generated based on the target risk level and health monitoring data, providing tailored prevention and response strategies for users at different risk levels. Early warning notifications based on the target risk level provide users with timely awareness of their skin microvascular inflammation risk, enabling them to take preventive or control measures before inflammation develops or in its early stages, effectively increasing their awareness of their health.

[0048] In order to make it easier for those skilled in the art to understand the present invention, Figure 2 According to a specific embodiment of the present invention, a method for predicting skin microvascular inflammation is described. Figure 2 As shown, the method for predicting skin microvascular inflammation includes: S210, obtaining the user's health monitoring data, and preprocessing the health monitoring data to obtain the preprocessed health monitoring data, where the health monitoring data includes skin image data, user health data, user current environment data, and user physiological parameters.

[0049] In the embodiment of the present invention, the implementation of step S210 may refer to the implementation of the above-mentioned step S110, and the present invention will not elaborate on this.

[0050] S220, obtaining a first eigenvector of the preprocessed skin image data, a second eigenvector of the user's health data, and a third eigenvector of the user's current environment data and the user's physiological parameters.

[0051] S230: Fusing the first eigenvector, the second eigenvector, and the third eigenvector to obtain a target eigenvector.

[0052] For example, the first eigenvector , the second eigenvector , the third eigenvector , among which, based on , and obtain the target feature vector.

[0053] In one embodiment of the present invention, the , calculate the weight of the first eigenvector; based on , calculate the weight of the second eigenvector; based on , calculate the weight of the third eigenvector, where represents the first eigenvector, represents the weight vector corresponding to the first eigenvector, represents the bias vector of the first eigenvector, represents the second eigenvector, represents the weight vector corresponding to the second eigenvector, represents the bias vector of the second eigenvector, represents the third eigenvector, represents the weight vector corresponding to the third eigenvector, Represents the bias vector of the third eigenvector, T represents the transpose operation of the matrix (i.e., linear algebra related), and exp represents the activation function (activation function selection includes but is not limited to ReLU, Tanh, Sigmoid, etc.); among them, the bias vector of the first eigenvector, the bias vector of the second eigenvector, and the bias vector of the third eigenvector, that is, the bias vector of each eigenvector, refers to the variable that needs to be added for the layer neurons in deep learning, and is also a kind of weight and can be trained; based on , perform weighted summation on the weight of the first eigenvector, the weight of the second eigenvector, and the weight of the third eigenvector to obtain the target eigenvector.

[0054] It should be noted that 、 and If overfitting occurs during training, you can add Regular or other constraints: ,in, Represents the first feature, Represents the second feature, Represents the third feature. After training, it can be analyzed 、 、 The distribution of 、 、 The numerical value of is used to evaluate the contribution of different modal features to the model decision. The fused feature vector It is then input into the subsequent deep learning model to complete the calculation of the inflammation risk score.

[0055] S240, input the target feature vector into the deep learning model, and output the user's first inflammation risk score.

[0056] In an embodiment of the present invention, the target feature vector is input into multiple attention heads in the deep learning model, and the multiple attention heads generate a query vector Q, a key vector K, and a value vector V by performing a linear transformation on the target feature vector; according to , get the attention weight of each attention head, where T represents the length of the data sequence, Represents the dimension of the key vector; according to , the attention weights of each attention head are concatenated and linearly transformed to obtain the fusion information, where h represents the number of attention heads, Represents the output transformation matrix; based on , the first calculation of residual connection and layer normalization, where Z represents the fusion information, Represents the input data sequence; based on , calculate the feedforward network, where W1, W2, 、 To learn the parameters, Represents the characteristic matrix during the operation; based on , calculate the residual connection and layer normalization again, where, Represents the encoder output of the current layer; based on , obtain the target features output by the multi-layer encoder; input the target features into the fully connected layer in the deep learning model to obtain the user's first inflammation risk score, where F represents the fused feature vector.

[0057] Key components of the Transformer architecture include residual connections and layer normalization. Residual connections (the Add component) are designed to address the challenges of training deep neural networks. By directly passing a portion of information from the previous layer to the next, residual connections ensure that at least some information from the previous layer is retained, improving model performance even when the performance of the next layer is uncertain. This strategy has been widely used in image processing, such as in ResNet.

[0058] Layer normalization (the Norm part) accelerates model training and promotes rapid convergence by normalizing layer activations. Compared to batch normalization (BN), layer normalization (LN) operates on different features of the same sample, while BN operates on the same feature across different samples. Although similar, LN is more applicable in the field of natural language processing (NLP) because it scales a sentence and typically processes word vectors, with consistent feature dimensions across all features. This avoids scaling issues that arise in BN due to varying feature dimensions.

[0059] S250: Determine the range of the dynamic threshold.

[0060] In an embodiment of the present invention, based on , calculate the risk offset, where Q represents the risk offset, represents the risk contribution value of each user feature, The weight of each user characteristic, including health data, current environmental data, and physiological parameters, is represented. The sum of the preset initial threshold and the risk offset is used as the dynamic threshold. The range of the dynamic threshold is determined based on the dynamic threshold and the fluctuation. For example, Δ represents the fluctuation, and the range of the dynamic threshold is [dynamic threshold - Δ, dynamic threshold + Δ].

[0061] In one embodiment of the present invention, it is also possible to , calculate the risk offset, where n represents the length of the input t vector (where the vector length needs to be specified according to the actual modeling), that is, the length of the risk contribution of the user feature, which is equivalent to the above f format, i and j represent the current cursor position, k represents the total length of the vector, W represents the weight of all user features, ω represents the trainable weight calculation, which can be compared with the above formula The format is equivalent, Represents manually specified non-trainable weights used to balance the deviations between multiple samples. represents the exponential decay factor, , represents the kth column of t, represents the j-th column of t, express The i-th column of represents the i-th column of ω; where, if , it means that the data length is insufficient and boundary processing can be performed appropriately.

[0062] It should be noted that It can be adjusted in real time according to the time span or data distribution changes. If the KL divergence (Kullback–Leibler Divergence) of the new data distribution compared to the historical distribution exceeds the threshold, it can be appropriately increased or decreased. This allows for faster or slower "memory decay." When the external environment experiences significant changes (such as sudden changes in temperature and humidity due to seasonal changes), dynamically adjusting the decay factor allows the model to more sensitively capture recent fluctuations.

[0063] based on , calculate the dynamic threshold, Indicates the preset initial threshold, represents the sum of the thresholds of multiple target risks, , Indicates the threshold for calculating target risk 1, Indicates the threshold for calculating target risk 2, Indicates the threshold for calculating target risk 3, Indicates the threshold for calculating the target risk n; the range of the dynamic threshold is determined based on the dynamic threshold and the upper and lower fluctuation amounts.

[0064] Among them, multiple target risks can be understood as the detected risks including multiple risks, and risks are considered to exist if they exceed the dynamic threshold.

[0065] S260: Determine a first risk level of the user based on the first inflammation risk score and the range of the dynamic threshold.

[0066] In an embodiment of the present invention, when it is judged that the first inflammation risk score is lower than the lower limit of the dynamic threshold range, the user's first risk level is determined to be low risk; when it is judged that the first inflammation risk score is within the range of the dynamic threshold, the user's first risk level is determined to be medium risk; when it is judged that the first inflammation risk score is higher than the upper limit of the dynamic threshold range, the user's first risk level is determined to be high risk.

[0067] In an embodiment of the present invention, the first risk level in the short term may also be predicted based on the acquired health impact parameters and historical health impact parameters of the user.

[0068] In embodiments of the present invention, a long-term first risk level can also be predicted based on the user's health history data and current environmental data. This uses a time series model (such as LSTM or Transformer-XL) to predict long-term trends and provide preventive recommendations.

[0069] S270 , when the trigger rule conditions are met, verify the first risk level based on the user's health impact parameter to obtain the user's target risk level.

[0070] In the embodiment of the present invention, the trigger rule conditions include environmental influence rules, medical history influence rules, and comprehensive factor influence rules.

[0071] Among them, the environmental impact rule defines that if the user's ambient temperature has increased significantly and the microvascular density has decreased recently, the inflammation risk level will be increased; the medical history impact rule defines that if the user has a history of diabetes and the microvascular morphology is abnormal, the inflammation risk level will be increased; the comprehensive factor rule defines that a combination of multiple health data (such as age, medical history) and environmental parameters is used to determine whether to increase the inflammation risk level. For example, if the user is young, has no medical history, and the corresponding values of the environmental parameters are all within the preset range, the inflammation risk level will be reduced, otherwise the inflammation risk level will be increased.

[0072] In the embodiment of the present invention, when any one of the triggering rule conditions is met, the user's health impact parameter is obtained; under the user's health impact parameter, based on , calculate the adjustment coefficient for the first risk level, where represents the scaling factor, A represents the result of nonlinear activation of the original score, , S represents the original score, , a+b+c represents the total number of items, which means the number of input variables to be considered, and m represents the index variable to be summed. represents the weight coefficient of the input variable (the weight coefficient controls the proportion or importance of each input variable in the final calculation. Different weight values will affect the final size of the original score S, thereby adjusting the contribution of each variable), Represents the input variable; based on the adjustment coefficient of the first risk level, the first risk level is used as the target risk level, or the first risk level is increased and the increased first risk level is used as the target risk level, or the first risk level is reduced and the reduced first risk level is used as the target risk level.

[0073] Among them, let the health impact parameter set be , a trainable or preset weight vector of winning bets Among them, In the model, negative factors (such as a history of diabetes and high blood lipids) can be given positive weights, and positive factors (such as healthy exercise habits and anti-inflammatory diet) can be given negative weights, thereby realizing a differentiated mechanism of "raising or lowering" the risk level when calculating the original score S.

[0074] In one embodiment of the present invention, when the adjustment coefficient of the first risk level is obtained, the adjustment coefficient of the first risk level can be compared with multiple thresholds to determine whether to increase or decrease the first risk level. In the case of In the case of In the case of , the first risk level remains unchanged; if In the case of In the case of the first risk level, the first risk level can be increased by two levels, among which, Indicates the first risk level, and Both represent thresholds, and the increased or decreased risk level is used as the target risk level.

[0075] Among them, the user's health influencing parameters include but are not limited to user physiological parameters, user environmental parameters, user age, user medical history, etc.

[0076] In an embodiment of the present invention, when multiple trigger rule conditions are met at the same time, the weight value of each trigger rule condition is obtained; based on the weight value of each trigger rule condition, the risk level under each trigger rule condition is obtained; and the sum of the risk levels under each trigger rule condition is used as the target risk level.

[0077] The weight value of each trigger rule condition may be pre-set or dynamically calculated according to an algorithm or rule.

[0078] That is to say, when multiple trigger rule conditions are met at the same time, the risk level corresponding to each trigger rule condition can be obtained, wherein the risk level corresponding to each trigger rule condition can refer to the above-mentioned implementation method of determining the risk level when any trigger rule condition is met, and then Determine the target risk level, where represents the risk level under the i-th trigger rule condition, Indicates the weight value of the i-th trigger rule condition.

[0079] In one embodiment of the present invention, when multiple trigger rule conditions are met at the same time, the weight value of each trigger rule condition is obtained; based on , calculate the target risk level, where h represents the number of trigger rule conditions, Indicates the triggering rule conditions The weight value of Indicates the triggering rule conditions The confidence level of the current user feature, Indicates mapping discrete levels to numerical values, i and j represent the current cursor position, Indicates the interaction coefficient of multiple trigger rule conditions (the interaction coefficient is preset or customized based on expert experience). Indicates the interaction confidence of multiple trigger rule conditions (the interaction confidence is based on expert experience and is a value between 0 and 1, where 0 indicates no confidence and 1 indicates full trust). Indicates the similarity between multiple trigger rule conditions.

[0080] in, Indicates the triggering rule conditions The confidence level of the current user feature. Do softmax calculation to get, that is , where K represents the total number of features, represents the i-th feature, express The exponential function of , ensures that the output is positive. In this case, the triggering rule conditions have little impact on the composite grade; on the contrary, if the confidence is high and consistent with other rules pointing to high risk, the final grade may increase significantly. Discrete levels refer to converting variables (risk levels) into discrete categories or levels (e.g., 1 = low risk, 2 = medium risk, 3 = high risk, converted to numerical values, e.g., low risk is mapped to 0.2, medium risk is mapped to 0.5, and high risk is mapped to 0.8).

[0081] in, Represents the similarity between multiple trigger rule conditions, where ,in, represents the i-th feature, represents the jth feature.

[0082] S280, providing early warning notifications based on the target risk level and generating personalized recommendations based on the target risk level and health monitoring data.

[0083] In an embodiment of the present invention, when the target risk level is determined to be high risk, an early warning notification can be automatically triggered, wherein the user can be reminded based on a variety of notification methods, such as mobile applications, text messages or emails, to ensure that the user can receive the early warning information in a timely manner and take necessary measures.

[0084] In an embodiment of the present invention, the personalized suggestions include diet adjustment, exercise suggestions, environmental improvement and medical examination suggestions.

[0085] Among them, dietary adjustments can be understood as recommendations based on health status and risk analysis to increase anti-inflammatory or antioxidant foods (such as green leafy vegetables, nuts) and reduce high-sugar, high-fat diets to reduce the risk of inflammation; exercise recommendations can be understood as combining the user's physical fitness and health data to recommend appropriate exercise types and frequencies (such as 30 minutes of light aerobic exercise per day) to improve blood circulation and microvascular health; environmental improvements can be understood as dynamic adjustment recommendations. For example, for low-humidity environments, it is recommended to use a humidifier or adjust the room temperature to a humidity range of 40%-60% to avoid harsh environments that aggravate skin symptoms. Medical examination recommendations can be understood as regular monitoring of microvascular changes or timely consultation with a doctor.

[0086] In an embodiment of the present invention, personalized recommendations can be generated by integrating user health data, environmental parameters, and historical feedback based on a Transformer-based generative model.

[0087] In an embodiment of the present invention, an adaptive learning mechanism is also included, including online learning and personalized adjustment.

[0088] Online learning is implemented as follows: Online learning receives and processes new data in real time using clear criteria. The system uses Wasserstein distance or KL divergence to compare the new data with the existing data distribution. If the distribution difference detected exceeds the set threshold, the system immediately triggers the update of the deep learning model parameters. Distribution changes in time series are dynamically evaluated by a sliding window statistical strategy to ensure the sensitivity of the deep learning model to data change trends. The system monitors the prediction error in real time and calculates the performance of the deep learning model based on mean square error or cross entropy. When the error exceeds the preset threshold, online training is initiated to optimize the predictive ability of the deep learning model. In addition, the system uses isolation forest or one-class support vector machine for novelty detection. When unseen data patterns are discovered, the deep learning model is updated first to ensure adaptability. Online training uses online gradient descent or replay-based incremental learning methods to perform local optimization only on the newly added data, avoiding full model retraining to improve efficiency.

[0089] Personalized adjustment is implemented as follows: Personalized adjustment builds a clustering model based on the characteristics of the user's health data. The system uses the K-means algorithm to divide users into multiple cluster groups, and trains a sub-model for each group separately to reflect the specific characteristics of the group. When a new user joins, the system calculates the similarity with the cluster center, matches the most appropriate sub-model for prediction, and fine-tunes the parameters in real time during use. The system trains the global model through the Model-Agnostic Meta-Learning framework, enabling it to quickly adapt to the data of a single user. In each personalized adjustment, the model fine-tunes the weights with a small sample to ensure that the needs of individual users are met. In addition, the system dynamically adjusts the distribution of feature weights. For users with special medical histories such as diabetes, the system automatically increases the weight of microvascular morphological features to make the prediction results more accurate and reliable.

[0090] Through the description of the above embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0091] According to one aspect of the embodiments of the present invention, a skin microvascular inflammation prediction system is also proposed. Figure 3 is a structural block diagram of a skin microvascular inflammation prediction system according to an embodiment of the present invention; Figure 3 As shown, including: An acquisition module 310 is configured to acquire health monitoring data of a user and preprocess the health monitoring data to obtain the preprocessed health monitoring data, wherein the health monitoring data includes skin image data, user health data, user current environment data, and user physiological parameters; A first obtaining module 320 is configured to predict the pre-processed health monitoring data based on a deep learning model to obtain a first inflammation risk score of the user; a determination module 330, configured to determine a first risk level of the user based on the first inflammation risk score and a range of a dynamic threshold, where the first risk level includes low risk, medium risk, and high risk; The second obtaining module 340 is used to verify the first risk level based on the user's health impact parameters when the trigger rule conditions are met, obtain the user's target risk level, and issue an early warning notification based on the target risk level and generate personalized suggestions based on the target risk level and the health monitoring data.

[0092] According to the skin microvascular inflammation prediction system of the embodiment of the present invention, the user's health monitoring data is obtained and preprocessed to obtain the preprocessed health monitoring data, which includes skin image data, user health data, user current environment data and user physiological parameters; the preprocessed health monitoring data is predicted based on the deep learning model to obtain the user's first inflammation risk score; based on the first inflammation risk score and the range of the dynamic threshold, the user's first risk level is determined, and the first risk level includes low risk, medium risk and high risk; when the trigger rule conditions are met, the first risk level is verified based on the user's health impact parameters to obtain the user's target risk level, and an early warning notification is issued based on the target risk level, and personalized suggestions are generated based on the target risk level and health monitoring data. In this way, skin microvascular inflammation can be accurately predicted.

[0093] Optionally, the first obtaining module 320 is specifically used to obtain the first eigenvector of the preprocessed skin image data, the second eigenvector of the user health data, and the third eigenvector of the user's current environment data and the user's physiological parameters; fuse the first eigenvector, the second eigenvector and the third eigenvector to obtain a target eigenvector; input the target eigenvector into the deep learning model, and output the first inflammation risk score of the user.

[0094] Optionally, the first obtaining module 320 is specifically configured to: , calculate the weight of the first eigenvector; based on , calculate the weight of the second eigenvector; based on , calculate the weight of the third eigenvector, where represents the first eigenvector, represents the weight vector corresponding to the first eigenvector, represents the bias vector of the first eigenvector, represents the second eigenvector, represents the weight vector corresponding to the second eigenvector, represents the bias vector of the second eigenvector, represents the third eigenvector, represents the weight vector corresponding to the third eigenvector, Represents the bias vector of the third eigenvector, T represents the transpose operation of the matrix, and exp represents the activation function; based on , performing weighted summation on the weight of the first feature vector, the weight of the second feature vector, and the weight of the third feature vector to obtain the target feature vector; wherein the target feature vector is input into multiple attention heads in the deep learning model, and the multiple attention heads generate a query vector Q, a key vector K, and a value vector V by performing a linear transformation on the target feature vector; according to , get the attention weight of each attention head, where T represents the length of the data sequence, Represents the dimension of the key vector; according to , the attention weights of each attention head are concatenated and linearly transformed to obtain fusion information, where h represents the number of attention heads, Represents the output transformation matrix; based on , the first calculation of residual connection and layer normalization, where Z represents the fusion information, Represents the input data sequence; based on , calculate the feedforward network, where W1, W2, 、 To learn the parameters, Represents the characteristic matrix during the operation; based on , calculate the residual connection and layer normalization again, where, Represents the encoder output of the current layer; based on , obtaining the target features output by the multi-layer encoder; inputting the target features into the fully connected layer in the deep learning model to obtain the first inflammation risk score of the user, where F represents the fused feature vector.

[0095] Optionally, determining the range of the dynamic threshold includes: based on , calculate the risk offset, where Q represents the risk offset, represents the risk contribution value of each user feature, Represents the weight of each user feature, wherein the user features include user health data, current environment data and physiological parameters; the sum of the preset initial threshold and the risk offset is used as the dynamic threshold; the range of the dynamic threshold is determined based on the dynamic threshold and the upper and lower fluctuations, or based on , calculate the risk offset, where n represents the length of the input t vector, i and j represent the current cursor positions, k represents the total length of the vector, W represents the weight of all user features, and ω represents the trainable weight matrix. Represents manually specified non-trainable weights used to balance the deviations between multiple samples. represents the exponential decay factor, , represents the kth column of t, represents the j-th column of t, express The i-th column of represents the i-th column of ω; based on , calculate the dynamic threshold, Indicates the preset initial threshold, represents the sum of the thresholds of multiple target risks, , Indicates the threshold for calculating target risk 1, Indicates the threshold for calculating target risk 2, Indicates the threshold for calculating target risk 3, Represents a threshold for calculating the target risk n; determining the range of the dynamic threshold based on the dynamic threshold and the upper and lower fluctuation amounts; wherein, the determination module 330 is specifically used to determine that the first risk level of the user is low risk when it is judged that the first inflammation risk score is lower than the lower limit of the range of the dynamic threshold; determine that the first risk level of the user is medium risk when it is judged that the first inflammation risk score is within the range of the dynamic threshold; determine that the first risk level of the user is high risk when it is judged that the first inflammation risk score is higher than the upper limit of the range of the dynamic threshold.

[0096] Optionally, the trigger rule conditions include environmental influence rules, medical history influence rules, and comprehensive factor influence rules, wherein the second obtaining module 340 is specifically used to obtain the health influence parameters of the user when any one of the trigger rule conditions is met; under the health influence parameters of the user, based on , calculate the adjustment coefficient for the first risk level, where represents the scaling factor, A represents the result of nonlinear activation of the original score, , S represents the original score, , a+b+c represents the total number of items, which means the number of input variables to be considered, and m represents the index variable to be summed. represents the weight coefficient of the input variable, represents an input variable; based on the adjustment coefficient of the first risk level, the first risk level is used as the target risk level, or the first risk level is increased and the increased first risk level is used as the target risk level, or the first risk level is reduced and the reduced first risk level is used as the target risk level.

[0097] Optionally, the second obtaining module 340 is specifically used to obtain the weight value of each trigger rule condition when multiple trigger rule conditions are met at the same time; obtain the risk level under each trigger rule condition based on the weight value of each trigger rule condition; use the sum of the risk levels under each trigger rule condition as the target risk level; or obtain the weight value of each trigger rule condition when multiple trigger rule conditions are met at the same time; based on , calculate the target risk level, where h represents the number of trigger rule conditions, Indicates the triggering rule conditions The weight value of Indicates the triggering rule conditions The confidence level of the current user feature, Indicates mapping discrete levels to numerical values, i and j represent the current cursor position, Represents the interaction coefficient of multiple trigger rule conditions, Indicates the interactive confidence of multiple trigger rule conditions, Indicates the similarity between multiple trigger rule conditions.

[0098] Optionally, the personalized suggestions include diet adjustments, exercise suggestions, environmental improvements, and medical examination suggestions.

[0099] According to one aspect of an embodiment of the present invention, an electronic device is provided.

[0100] Figure 4 FIG. 1 is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Figure 4 As shown, the electronic device may include one or more ( Figure 4 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a microprocessor (Microprocessor Unit, referred to as MPU) or a programmable logic device (Programmable logic device, referred to as PLD)) and a memory 104 for storing data. In an exemplary embodiment, the electronic device may further include a transmission device 106 and an input / output device 108 for communication functions. It will be understood by those skilled in the art that Figure 4 The structure shown is only for illustration and does not limit the structure of the above terminal device. Figure 4 More or fewer components than shown, or with Figure 4 Equivalent functions or comparisons shown Figure 4 Shown are different configurations with more functionality.

[0101] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the method for predicting skin microvascular inflammation in the embodiments of the present invention. The processor 102 executes the computer program stored in the memory 104 to perform various functional applications and data processing, thereby implementing the above-mentioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor 102, and such remote memory may be connected to the terminal device via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0102] Transmission device 106 is used to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by a communications provider of a switching device. In one embodiment, transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0103] The present invention provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute a skin microvascular inflammation prediction method.

[0104] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation plans of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, and is not a limitation on the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

[0105] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0106] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and are not to be construed as limiting the present invention. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0107] Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they may still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein with equivalents; and such modifications or replacements do not deviate from the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for predicting skin microvascular inflammation, characterized in that: include: S1, obtaining health monitoring data of a user and preprocessing the health monitoring data to obtain the preprocessed health monitoring data, wherein the health monitoring data includes skin image data, user health data, user current environment data, and user physiological parameters; S2, predicting the pre-processed health monitoring data based on a deep learning model to obtain a first inflammation risk score for the user; S3, determining a first risk level of the user based on the first inflammation risk score and a range of a dynamic threshold, where the first risk level includes low risk, medium risk, and high risk; S4. When the trigger rule conditions are met, the first risk level is verified based on the user's health impact parameters to obtain the user's target risk level, and an early warning notification is issued based on the target risk level and personalized recommendations are generated based on the target risk level and the health monitoring data.

2. The method for predicting skin microvascular inflammation according to claim 1, wherein: Predicting the preprocessed health monitoring data based on a deep learning model to obtain a first inflammation risk score for the user includes: Obtaining a first eigenvector of the preprocessed skin image data, a second eigenvector of the user health data, and a third eigenvector of the user's current environment data and the user's physiological parameters; Fusing the first eigenvector, the second eigenvector, and the third eigenvector to obtain a target eigenvector; The target feature vector is input into the deep learning model, and the first inflammation risk score of the user is output.

3. The method for predicting skin microvascular inflammation according to claim 2, characterized in that: Fusing the first eigenvector, the second eigenvector, and the third eigenvector to obtain a target eigenvector includes: based on , calculate the weight of the first eigenvector; based on , calculate the weight of the second eigenvector; based on , calculate the weight of the third eigenvector, where represents the first eigenvector, represents the weight vector corresponding to the first eigenvector, represents the bias vector of the first eigenvector, represents the second eigenvector, represents the weight vector corresponding to the second eigenvector, represents the bias vector of the second eigenvector, represents the third eigenvector, represents the weight vector corresponding to the third eigenvector, represents the bias vector of the third eigenvector, T represents the transpose operation of the matrix, and exp represents the activation function; based on , performing weighted summation on the weight of the first eigenvector, the weight of the second eigenvector, and the weight of the third eigenvector to obtain the target eigenvector; Inputting the target feature vector into the deep learning model and outputting the first inflammation risk score of the user includes: Input the target feature vector into multiple attention heads in the deep learning model, and the multiple attention heads generate a query vector Q, a key vector K, and a value vector V by performing a linear transformation on the target feature vector; according to , get the attention weight of each attention head, where T represents the length of the data sequence, represents the dimension of the key vector; according to , the attention weights of each attention head are concatenated and linearly transformed to obtain fusion information, where h represents the number of attention heads, W o represents the output transformation matrix; based on , the first calculation of residual connection and layer normalization, where Z represents the fusion information, Represents the input data sequence; based on , calculate the feedforward network, where W1, W2, 、 To learn the parameters, Represents the characteristic matrix during the operation; based on , calculate the residual connection and layer normalization again, where, Represents the encoder output of the current layer; based on , get the target features output by the multi-layer encoder; The target feature is input into the fully connected layer in the deep learning model to obtain the first inflammation risk score of the user, where F represents the fused feature vector.

4. The method for predicting skin microvascular inflammation according to claim 1, wherein: Determining the range of the dynamic threshold includes: based on , calculate the risk offset, where Q represents the risk offset, represents the risk contribution value of each user feature, Represents the weight of each user feature, where user features include user health data, current environment data, and physiological parameters; The sum of the preset initial threshold and the risk offset is used as the dynamic threshold; Determine the range of the dynamic threshold based on the dynamic threshold and the up and down fluctuation amount, or based on , calculate the risk offset, where n represents the length of the input t vector, i and j represent the current cursor positions, k represents the total length of the vector, W represents the weight of all user features, and ω represents the trainable weight calculation. Represents manually specified non-trainable weights used to balance the deviations between multiple samples. represents the exponential decay factor, , represents the kth column of t, represents the j-th column of t, express The i-th column of represents the i-th column of ω; based on , calculate the dynamic threshold, Indicates the preset initial threshold, represents the sum of the thresholds of multiple target risks, , Indicates the threshold for calculating target risk 1, Indicates the threshold for calculating target risk 2, Indicates the threshold for calculating target risk 3, Indicates the threshold for calculating the target risk n; Determining a range of the dynamic threshold based on the dynamic threshold and the upper and lower fluctuation amounts; Determining the first risk level of the user based on the first inflammation risk score and the range of the dynamic threshold includes: If it is determined that the first inflammation risk score is lower than a lower limit of the dynamic threshold, determining that the first risk level of the user is low risk; If it is determined that the first inflammation risk score is within the range of the dynamic threshold, determining the first risk level of the user as medium risk; When it is determined that the first inflammation risk score is higher than the upper limit of the dynamic threshold range, the first risk level of the user is determined to be high risk.

5. The method for predicting skin microvascular inflammation according to claim 1, wherein: The trigger rule conditions include an environmental impact rule, a medical history impact rule, and a comprehensive factor impact rule. When the trigger rule conditions are met, the first risk level is verified based on the user's health impact parameters to obtain the user's target risk level, including: When any one of the trigger rule conditions is met, obtaining the health impact parameter of the user; Under the user's health impact parameters, based on , calculate the adjustment coefficient for the first risk level, where represents the scaling factor, A represents the result of nonlinear activation of the original score, , S represents the original score, , a+b+c represents the number of input variables to be considered, m represents the index variable to be summed, represents the weight coefficient of the input variable, represents the input variable; Based on the adjustment coefficient of the first risk level, the first risk level is used as the target risk level, or the first risk level is increased and the increased first risk level is used as the target risk level, or the first risk level is reduced and the reduced first risk level is used as the target risk level.

6. The method for predicting skin microvascular inflammation according to claim 5, characterized in that: Also includes: When multiple trigger rule conditions are met simultaneously, obtaining a weight value of each trigger rule condition; Based on the weight value of each trigger rule condition, obtaining the risk level under each trigger rule condition; The sum of the risk levels under each of the triggering rule conditions is used as the target risk level; or When multiple trigger rule conditions are met simultaneously, obtaining a weight value of each trigger rule condition; based on , calculate the target risk level, where h represents the number of trigger rule conditions, Indicates the triggering rule conditions The weight value of Indicates the triggering rule conditions The confidence level of the current user feature, Indicates mapping discrete levels to numerical values, i and j represent the current cursor position, Represents the interaction coefficient of multiple trigger rule conditions, Indicates the interactive confidence of multiple trigger rule conditions, Indicates the similarity between multiple trigger rule conditions.

7. The method for predicting skin microvascular inflammation according to claim 1, characterized in that: The personalized recommendations include dietary adjustments, exercise recommendations, environmental improvements, and medical examination recommendations.

8. A skin microvascular inflammation prediction system, characterized in that: include: an acquisition module, configured to acquire the user's health monitoring data and preprocess the health monitoring data to obtain the preprocessed health monitoring data, wherein the health monitoring data includes skin image data, user health data, user current environment data, and user physiological parameters; A first obtaining module is configured to predict the pre-processed health monitoring data based on a deep learning model to obtain a first inflammation risk score of the user; a determination module, configured to determine a first risk level of the user based on the first inflammation risk score and a range of a dynamic threshold, wherein the first risk level includes low risk, medium risk, and high risk; The second obtaining module is used to verify the first risk level based on the user's health impact parameters when the trigger rule conditions are met, obtain the user's target risk level, issue an early warning notification based on the target risk level, and generate personalized suggestions based on the target risk level and the health monitoring data.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 7.

Images