Climacteric syndrome health management platform based on big data

Abstract

The invention discloses a climacteric syndrome health management platform based on big data, and relates to the field of female health management. Comprising a marker monitoring module, a data acquisition and processing module, a health portrait modeling module, a personal health assessment module, a climacteric prediction module, an intervention scheme recommendation module, a medical data protection module, a digital health interaction module, a combined intervention module, a health community mutual assistance module and an exercise guidance module. According to the method, more accurate climacteric health state prediction and trend visualization are realized, the evaluation precision of climacteric symptom severity is improved, manual intervention strategy adjustment is not needed, the intervention effect is improved, the user compliance is enhanced, data security and privacy protection are ensured, and the user experience is improved by combining the physique and physiological indexes of the user. A personalized traditional Chinese and western medicine combined intervention scheme is matched, and the scientificity and effectiveness of intervention are improved.

Description

Technical Field

[0001] The present invention relates to the field of women's health management, and in particular to a menopausal syndrome health management platform based on big data. Background Art

[0002] Menopausal Syndrome is a major health issue that affects hundreds of millions of women worldwide. Menopause is an important stage in a woman's life cycle, usually occurring between the ages of 45 and 55. Its symptoms involve multiple aspects such as endocrine, nervous system, metabolism, and mental health. Along with a drop in estrogen levels, it triggers a series of physiological and psychological symptoms, such as hot flashes, night sweats, anxiety, insomnia, osteoporosis, and increased risk of cardiovascular disease. Traditional health management methods mainly rely on regular physical examinations, questionnaire assessments, or single physiological indicator tests. They cannot achieve comprehensive and dynamic monitoring of the health status of menopausal women, and lack personalized intervention strategies. Therefore, the development of a menopausal syndrome health management platform based on big data is of great significance for improving the quality of life of menopausal women and reducing the risk of chronic diseases.

[0003] The existing menopausal syndrome health management platform cannot achieve accurate prediction of menopausal health status and trend visualization, which reduces the accuracy of assessing the severity of menopausal symptoms, and cannot match personalized Chinese and Western medicine intervention plans based on user constitution and physiological indicators, which reduces the scientificity and effectiveness of the intervention. To this end, we propose a menopausal syndrome health management platform based on big data. Summary of the invention

[0004] The purpose of the present invention is to solve the above-mentioned problems and to provide a menopausal syndrome health management platform based on big data.

[0005] The present invention proposes a menopausal syndrome health management platform based on big data, the system comprising: a marker monitoring module, a data acquisition and processing module, a health portrait modeling module, a personal health assessment module, a menopausal prediction module, an intervention program recommendation module, a medical data protection module, a digital health interaction module, a combined intervention module, a health community mutual assistance module and an exercise guidance module;

[0006] The marker monitoring module uses home testing equipment to mine menopause-specific biomarkers;

[0007] The data acquisition and processing module integrates and processes physiological indicators, wearable device data, behavioral data, environmental data, psychological assessment and genetic information to construct a menopause-specific data set;

[0008] The health portrait modeling module dynamically builds a personalized user health portrait based on the menopause-specific data set;

[0009] The personal health assessment module combines the user's real-time data to assess the severity of the user's menopausal symptoms;

[0010] The menopause prediction module predicts the future menopause symptoms of the user and estimates their severity based on the user's health profile and the evaluation results of menopause symptoms;

[0011] The intervention plan recommendation module dynamically optimizes the intervention plan based on user data and feedback;

[0012] The medical data protection module is used to encrypt the data collected by the data collection and processing module and locally update the parameters of the menopause prediction module;

[0013] The digital health interaction module is used to identify the user's emotional state and dynamically adjust the communication strategy;

[0014] The combined intervention module is used to associate traditional Chinese medicine theory with Western medicine biological indicators and recommend personalized integrated Chinese and Western medicine programs;

[0015] The health community mutual aid module is used to receive user health data actively uploaded by users and establish a patient mutual aid community;

[0016] The motion guidance module guides the user's motion in real time through skeleton key point detection.

[0017] As a further solution of the present invention, the specific steps of the data acquisition and processing module to construct a menopause-specific data set are as follows:

[0018] Q1.1: The data collection and processing module uses linear interpolation to time-align the data from different data sources, and then uses multiple interpolation to process the missing values ​​in each data set. After the data alignment and missing value processing are completed, the data from different data sources are arranged from small to large, and the arranged data from each data source are divided into four equal parts;

[0019] Q1.2: Take each quantile as the lower quartile Q1, median Q2 and upper quartile Q3, and obtain the corresponding interquartile range IQR through Q3-Q1. Then calculate the upper limit Q3+1.5IQR and lower limit Q1-1.5IQR of each type of data respectively. Then mark the data exceeding the upper and lower limits as outliers and remove them. Then use Kalman filtering to fuse the processed data to establish a menopause-specific data set.

[0020] As a further solution of the present invention, the specific steps of the health portrait modeling module dynamically constructing a personalized user health portrait are as follows:

[0021] S1.1: Arrange the data in each group of the menopause-specific dataset in chronological order, extract the characteristic information in each group of data through principal component analysis, standardize the characteristic information of different dimensions, calculate the characteristic contribution of each characteristic information, and integrate the processed characteristic information of each group to construct a feature set;

[0022] S1.2: Split the feature set into multiple groups of time series slices according to a preset fixed-length window, then add position codes to each time series slice, calculate the query, key, and value matrices of each group of time series slices, and independently calculate the query, key, and value matrices on multiple attention heads;

[0023] S1.3: Calculate the attention scores of each group of time series slices through the query matrix and the key matrix, and establish the attention score matrix. Then use the Softmax function to normalize the attention scores, and perform weighted summation on the processed attention scores of each group. The weighted summation result of each attention head is used as the final attention output.

[0024] S1.4: The final attention results output by each attention head are normalized through LayerNorm, and then the residuals are combined. After that, the combined attention values ​​are nonlinearly transformed through a fully connected neural network, and the Transformer-encoded time series feature matrix is ​​output. Each feature data in the time series feature matrix is ​​input into the LSTM model.

[0025] S1.5: The LSTM model processes each feature data through a gating mechanism, and outputs the health status prediction value of the corresponding user through the fully connected layer of the model. It calculates the Euclidean distance between data points at different times and uses it as the high-dimensional distance between each data point. It also uses Gaussian distribution to perform high-dimensional distribution modeling on the similarity between each data point, and uses Cauchy distribution to perform low-dimensional distribution modeling on each data point.

[0026] S1.6: The difference between the high-dimensional distribution and the low-dimensional distribution is calculated through the KL divergence function. Based on the difference values, the low-dimensional coordinates are updated through the gradient descent algorithm, and the corresponding 2D or 3D low-dimensional representation is generated, including different stages of menopausal health status, visualized health trends, and personalized health recommendations.

[0027] As a further solution of the present invention, the specific steps of the personal health assessment module to assess the severity of the user's menopausal symptoms are as follows:

[0028] S2.1: Collect medical knowledge data from various data sources including medical literature, medical guidelines, electronic health records, medical terminology, and clinical scoring systems, use NLP technology to convert unstructured medical literature into structured data, traverse each set of processed medical knowledge data, merge duplicate entries of the same disease, symptom, and treatment method in the data, and then map medical knowledge data from different sources into corresponding standard medical terms;

[0029] S2.2: Extract the entities of diseases, symptoms, physiological indicators, treatment plans, lifestyle factors and environmental factors from the processed medical knowledge data through NRE technology, identify the relationship between the entities, establish knowledge triples in the form of (head entity, relationship, tail entity), and store the triple data in GraphDB to establish the corresponding medical knowledge graph library;

[0030] S2.3: Collect information on menopausal symptoms in the medical knowledge graph and calculate the user's clinical score , if 0≤ ≤6, the user is judged to have no menopausal symptoms. ≤15, the user is judged to have mild menopausal symptoms. ≤30, the user is judged to have moderate menopausal symptoms. , the user is judged to have severe menopausal symptoms, among which the clinical score The specific calculation formula is as follows:

[0031] ;

[0032] In the formula, represents the menopausal symptom score, Represents the total number of symptoms, Representative The weight of the symptoms, Representative The severity of the symptoms;

[0033] S2.4: Based on the real-time collected physiological data, behavioral data, environmental data, and psychological state data of each user, the real-time data is mapped to the symptom severity correction factor through linear regression. Then, based on the symptom severity correction factor, the score of each menopausal symptom is corrected. Then, the similarity between symptoms is calculated based on the medical knowledge graph library, and the potential association between symptoms is inferred through logical rules. Then, the clinical score is calculated through K-Means. Perform final classification and output the severity of the user's final menopausal symptoms.

[0034] As a further solution of the present invention, the specific steps of the intervention plan recommendation module dynamically optimizing the intervention plan are as follows:

[0035] S3.1: Based on the multimodal data of users collected by the data acquisition and processing module, the corresponding health state space is constructed, and then the corresponding action space is established according to the intervention measures of adjusting hormone replacement therapy, recommending specific dietary adjustments, suggesting increasing aerobic exercise, recommending psychological intervention, and environmental optimization. The intervention plan is composed of different intervention measures. The reward function is established according to the symptom score, mental health score, user compliance change, and intervention cost to reflect the impact of the intervention plan on the health state;

[0036] S3.2: Decompose the intervention plan taken by the current user into an action set consisting of different intervention measures, calculate the intervention effect of the current intervention plan on the user's menopausal symptoms through the reward function, and then calculate the probability distribution of each intervention measure in the action space based on the current user's health status;

[0037] S3.3: At different intervention stages, the strategy network extracts multiple groups of intervention measures through the roulette wheel selection method based on the probability distribution of each intervention measure, and establishes corresponding intervention plans. Then, the intervention plans are executed, the changes in the health status of the user after the intervention are simulated and recorded, the intervention effect of the selected intervention plan is calculated through the reward function, and the intervention plans and their intervention effects constructed through multiple rounds of selection are recorded;

[0038] S3.4: The value network calculates the cumulative discounted rewards and temporal differences of each intervention plan at different stages, and calculates the advantage estimate of each intervention plan through the cumulative discounted rewards and temporal differences, and then calculates the probability change of the new intervention plan for the current action relative to the old intervention plan, and calculates the total loss value of the strategy network and the value network based on the advantage estimate and probability change;

[0039] S3.5: The Adam optimizer is used to optimize the parameters of the policy network and the value network. Then, strategy selection, advantage estimation, and parameter update are continuously performed until the change in the total loss value converges to the preset range. The iteration is stopped and the final generated intervention plan is output for real user intervention recommendations. User feedback is monitored in real time, and new data is recorded for subsequent intervention plan updates.

[0040] As a further solution of the present invention, the specific steps of the medical data protection module locally updating the parameters of the menopause prediction module are as follows:

[0041] S4.1: The medical data protection module initializes the global model parameters on the central server. At the same time, each medical institution establishes its local medical data set and local model respectively. The local medical data sets are not shared, and Paillier homomorphic encryption is used for local data.

[0042] S4.2: The central server sends the current global model parameters to each medical institution. The local model of each medical institution uses the current global model parameters as its own initial parameters and divides the local medical data set into a training set and a test set.

[0043] S4.3: Input the training set into the local model, the local model forward propagates the data in the training set, and outputs the prediction results of the user's menopausal symptoms. The loss value between the prediction result and the actual symptoms is calculated by the mean square error function. Then, based on the chain rule, the loss value is back-propagated from the output layer of the local model, and the gradient value of the loss value for the parameters of each network layer of the local model is calculated. At the same time, homomorphic encryption is used to protect the calculated gradient information;

[0044] S4.4: Each medical institution uploads the encrypted gradient information to the central server, which then sums the gradients of each gradient information in an encrypted state to obtain the corresponding global gradient. The central server decrypts the calculated global gradient and updates the global model parameters through the gradient descent algorithm, and then encrypts the updated global model parameters.

[0045] S4.5: The encrypted global model parameters are sent back to each medical institution. Each medical institution decrypts the model parameters and continues a new round of local training. The parameter sending, gradient calculation and global model parameter update are repeated until the loss value of the global model is lower than the preset threshold. The final global model parameters are used as the optimal parameters and sent to each medical institution. After that, the local model of each medical institution updates its own parameter information based on the optimal parameters and predicts the user's menopausal symptoms based on real-time user data.

[0046] As a further solution of the present invention, the specific steps of the combined intervention module to recommend a personalized integrated Chinese and Western medicine program are as follows:

[0047] S5.1: Collect the user's physical characteristics through the TCM physical constitution questionnaire and calculate the user's physical constitution score ,like If the value is higher than the preset constitution threshold, the user is judged to belong to this constitution type, where the constitution types include Yin deficiency, Yang deficiency, Qi deficiency, blood stasis and phlegm dampness.

[0048] S5.2: Extract the user's serum hormone level, cortisol level, bone density and metabolic parameters, and calculate the hormone imbalance index and stress index ,like If the level of hormone imbalance is higher than the preset threshold, it indicates that the user has a high degree of hormone imbalance. If the value is higher than the preset upper threshold, it indicates that the user is in a state of severe stress. If it is lower than the preset lower threshold, it indicates that the user has chronic fatigue syndrome;

[0049] S5.3: Use the expert knowledge base to establish a constitution-physiological index association table, and calculate the matching degree of different Chinese and Western medicine characteristics by weighted calculation, then select the constitution category with the highest matching degree as the constitution type corresponding to the user, and select the Chinese and Western medicine combination plan suitable for the user according to the user's constitution type, and combine it with the intervention plan generated by the intervention plan recommendation module to generate a personalized intervention plan;

[0050] S5.4: Calculate the personalized intervention score for each intervention plan. If the user feedback shows that the intervention effect is not good, optimize the plan through the intervention plan recommendation module and update the intervention strategy.

[0051] Beneficial effects of the present invention:

[0052] The present invention constructs a menopause-specific dataset by integrating and processing physiological indicators, wearable device data, behavioral data, environmental data, psychological assessment and genetic information, extracts features using principal component analysis, constructs time series data after standardization, extracts time series features through Transformer, and predicts health status in combination with LSTM. Subsequently, t-SNE was used to reduce the dimension and visualize health trends. NLP technology was then used to parse medical literature, guidelines, and electronic health records, extract entities such as diseases, symptoms, and treatment methods, and construct triple relationships. The severity of menopausal symptoms was evaluated in combination with a clinical scoring system. Reinforcement learning was used to optimize intervention strategies based on the predicted severity of users' menopausal symptoms. Personalized intervention plans were dynamically recommended based on information such as hormone levels and psychological state, and intervention strategies were optimized based on the PPO algorithm. After that, each institution, without sharing the original data, collaboratively trained a global model based on federated learning and homomorphic encryption to establish a constitution identification-Western medicine indicator linkage model to match personalized plans. Combined with intervention feedback optimization and adjustment, it ultimately provided a more accurate and adaptable health management plan, achieved more accurate prediction of menopausal health status and trend visualization, improved the accuracy of assessment of the severity of menopausal symptoms, and did not require manual intervention strategy adjustment, thereby improving the intervention effect and enhancing user compliance, ensuring data security and privacy protection, and matching personalized Chinese and Western medicine intervention plans based on user constitution and physiological indicators to improve the scientificity and effectiveness of the intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described below in conjunction with the accompanying drawings.

[0054] Figure 1 This is a framework diagram of a menopausal syndrome health management platform based on big data. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0057] Example 1

[0058] The embodiment of the present invention provides a menopausal syndrome health management platform based on big data. Figure 1 , Figure 1 A framework diagram of a menopausal syndrome health management platform based on big data provided by an embodiment of the present invention. The system includes:

[0059] Marker monitoring module, data collection and processing module, health portrait modeling module, personal health assessment module, menopause prediction module, intervention plan recommendation module, medical data protection module, digital health interaction module, combined intervention module, health community mutual assistance module and exercise guidance module.

[0060] The biomarker monitoring module uses home testing equipment to discover menopause-specific biomarkers.

[0061] The data acquisition and processing module integrates and processes physiological indicators, wearable device data, behavioral data, environmental data, psychological assessments, and genetic information to construct a menopause-specific data set.

[0062] Specifically, the data acquisition and processing module uses linear interpolation to time align the data sets collected from different data sources, and then uses multiple interpolation to process the missing values ​​in each data set. After the data alignment and missing value processing are completed, the data sets from different data sources are arranged from small to large, and the arranged data from each data source are divided into four equal parts. Each quartile is used as the lower quartile Q1, the median Q2, and the upper quartile Q3, and the corresponding interquartile range IQR is obtained through Q3-Q1. Then, the upper limit Q3+1.5IQR and the lower limit Q1-1.5IQR of each type of data are calculated respectively, and the data exceeding the upper and lower limits are marked as outliers and eliminated, and the processed data are fused using Kalman filtering to establish a menopause-specific data set.

[0063] The health portrait modeling module dynamically builds personalized user health portraits based on menopause-specific data sets.

[0064] Specifically, each group of data in the menopause-specific dataset is arranged in chronological order, and the characteristic information in each group of data is extracted by principal component analysis, and then the characteristic information of different dimensions is standardized, and the characteristic contribution of each characteristic information is calculated, and the processed characteristic information of each group is integrated to construct a feature set, and the feature set is divided into multiple groups of time series slices according to a preset fixed-length window, and then a position code is added to each time series slice, and the query, key and value matrices of each group of time series slices are calculated, and the query, key and value matrices are calculated independently on multiple attention heads, and the attention scores of each group of time series slices are calculated through the query matrix and the key matrix, and an attention score matrix is ​​established, and then the attention scores are normalized by the Softmax function, and the weighted sum of the processed attention scores of each group is performed, and the weighted sum result of each attention head is used as its final attention output, and the final attention result output by each attention head is processed by Layer rNorm is used for normalization, and then the residual combination is performed on the result. After that, the combined attention value is nonlinearly transformed through a fully connected neural network, and the time series feature matrix encoded by Transformer is output. The feature data in the time series feature matrix is ​​input into the LSTM model. The LSTM model processes each feature data through a gating mechanism, and outputs the corresponding user's health status prediction value through the fully connected layer of the model. The Euclidean distance between data points at different times is calculated and used as the high-dimensional distance between each data point. The high-dimensional distribution modeling of the similarity between each data point is performed through Gaussian distribution, and the low-dimensional distribution modeling of each data point is performed through Cauchy distribution. The difference value between the high-dimensional distribution and the low-dimensional distribution is calculated through the KL divergence function. Based on each difference value, the low-dimensional coordinates are updated through the gradient descent algorithm, and the corresponding 2D or 3D low-dimensional representation is generated, including different stages of menopausal health status, visualized health trends, and personalized health recommendations.

[0065] The personal health assessment module combines the user's real-time data to assess the severity of the user's menopausal symptoms.

[0066] Specifically, medical knowledge data are collected from various data sources such as medical literature, medical guidelines, electronic health records, medical terminology sets, and clinical scoring systems. NLP technology is used to convert unstructured medical literature into structured data. The processed groups of medical knowledge data are traversed, and repeated entries of the same disease, symptoms, and treatment methods in the data are merged. Medical knowledge data from different sources are then mapped to corresponding standard medical terms. Disease, symptoms, physiological indicators, treatment plans, lifestyle factors, and environmental factors are extracted from the processed medical knowledge data through NRE technology, and the relationship between each entity is identified to establish knowledge triples in the form of (head entity, relationship, tail entity). The corresponding medical knowledge graph library has been established by storing the triple data through GraphDB graph database, and the information on menopausal symptoms in the medical knowledge graph library is collected, and the user's clinical score is calculated. , if 0≤ ≤6, the user is judged to have no menopausal symptoms. ≤15, the user is judged to have mild menopausal symptoms. ≤30, the user is judged to have moderate menopausal symptoms. , the user is judged to have severe menopausal symptoms. According to the real-time collected physiological data, behavioral data, environmental data and psychological state data of each user, the real-time data is mapped to the symptom severity correction factor through linear regression. Then, based on the symptom severity correction factor, the score of each menopausal symptom is corrected. Then, the similarity between symptoms is calculated based on the medical knowledge graph library, and the potential association between symptoms is inferred through logical rules. Then, the clinical score is calculated through K-Means. Perform final classification and output the severity of the user's final menopausal symptoms.

[0067] It should be further explained that the clinical score The specific calculation formula is as follows:

[0068] ;

[0069] In the formula, represents the menopausal symptom score, Represents the total number of symptoms, Representative The weight of the symptoms, Representative The severity of the symptoms.

[0070] The menopause prediction module predicts the user's future menopausal symptoms and estimates their severity based on the user's health profile and menopausal symptom assessment results; the intervention plan recommendation module dynamically optimizes the intervention plan based on user data and feedback.

[0071] Specifically, based on the user multimodal data collected by the data acquisition and processing module, a corresponding health status space is constructed, and then a corresponding action space is established according to the intervention measures of adjusting hormone replacement therapy, recommending specific dietary adjustments, suggesting increasing aerobic exercise, recommending psychological intervention, and environmental optimization. The intervention plan is composed of different intervention measures. A reward function is established according to the symptom score, mental health score, user compliance change, and intervention cost to reflect the impact of the intervention plan on the health status. The intervention plan currently taken by the user is decomposed into an action set composed of different intervention measures. The intervention effect of the current intervention plan on the user's menopausal symptoms is calculated through the reward function. Then, the probability distribution of each intervention measure in the action space is calculated based on the current user's health status. At different intervention stages, the strategy network extracts multiple groups of intervention measures through the roulette selection method based on the probability distribution of each intervention measure, and establishes the corresponding intervention plan, which is then executed. The intervention plan is implemented, and the changes in the health status of the user after the intervention are simulated and recorded. The intervention effect of the selected intervention plan is calculated through the reward function, and the intervention plans constructed by multiple rounds of selection and their intervention effects are recorded. The value network calculates the cumulative discounted rewards and time-series differences of each intervention plan at different stages, and calculates the advantage estimate of each intervention plan through the cumulative discounted rewards and time-series differences. Then, the probability change of the new intervention plan for the current action relative to the old intervention plan is calculated, and the total loss value of the policy network and the value network is calculated according to the advantage estimate and probability change. The policy network and the value network are optimized by the Adam optimizer, and then the strategy selection, advantage estimation and parameter update are continuously performed until the change value of the total loss value converges to the preset range. The iteration is stopped, and the final generated intervention plan is output for real user intervention recommendation, and user feedback is monitored in real time, and new data is recorded for subsequent intervention plan updates.

[0072] Example 2

[0073] The embodiment of the present invention provides a menopausal syndrome health management platform based on big data. Figure 1 , Figure 1 A framework diagram of a menopausal syndrome health management platform based on big data provided by an embodiment of the present invention. The system includes:

[0074] Marker monitoring module, data collection and processing module, health portrait modeling module, personal health assessment module, menopause prediction module, intervention plan recommendation module, medical data protection module, digital health interaction module, combined intervention module, health community mutual assistance module and exercise guidance module.

[0075] The medical data protection module is used to encrypt the data collected by the data collection and processing module and locally update the parameters of the menopause prediction module.

[0076] Specifically, the medical data protection module initializes the global model parameters on the central server. At the same time, each medical institution establishes its local medical data set and local model respectively. The local medical data sets do not share data, and Paillier homomorphic encryption is used to encrypt local data. The central server sends the current global model parameters to each medical institution. The local model of each medical institution uses the current global model parameters as its own initial parameters, and divides the local medical data set into a training set and a test set. The training set is input into the local model. The local model forward propagates the data in the training set and outputs the prediction results of the user's menopausal symptoms. The loss value between the prediction result and the actual symptom is calculated by the mean square error function. Then, based on the chain rule, the loss value is back-propagated from the output layer of the local model, and the gradient of the loss value with respect to the parameters of each network layer of the local model is calculated. value, and homomorphic encryption is used to protect the calculated gradient information. Each medical institution uploads the encrypted gradient information to the central server, and then the central server sums the gradients of each gradient information in an encrypted state to obtain the corresponding global gradient. The central server decrypts the calculated global gradient and updates the global model parameters through the gradient descent algorithm. The updated global model parameters are then encrypted and the encrypted global model parameters are re-sent to each medical institution. Each medical institution decrypts the model parameters and continues a new round of local training, repeating parameter sending, gradient calculation, and global model parameter update until the loss value of the global model is lower than the preset threshold. The final global model parameters are used as the optimal parameters and sent to each medical institution. After that, the local model of each medical institution updates its own parameter information based on the optimal parameters, and predicts the user's menopausal symptoms based on real-time user data.

[0077] The digital health interaction module is used to identify the user's emotional state and dynamically adjust the communication strategy; the combined intervention module is used to associate traditional Chinese medicine theory with Western medicine biological indicators and recommend personalized traditional Chinese and Western medicine integration plans.

[0078] Specifically, the user's physical characteristics are collected through a TCM physical questionnaire, and the user's physical score is calculated. ,like If the physical condition is higher than the preset threshold, the user is judged to belong to this physical type, where physical types include Yin deficiency, Yang deficiency, Qi deficiency, blood stasis and phlegm dampness. The user's serum hormone level, cortisol level, bone density and metabolic parameters are extracted, and the hormone imbalance index is calculated. and stress index ,like If the level of hormone imbalance is higher than the preset threshold, it indicates that the user has a high degree of hormone imbalance. If the value is higher than the preset upper threshold, it indicates that the user is in a state of severe stress. If the value is lower than the preset lower threshold, it indicates that the user has chronic fatigue syndrome. The expert knowledge base is used to establish a constitution-physiological indicator association table, and the degree of matching of different Chinese and Western medicine characteristics is calculated by weighted calculation. Then, the constitution category with the highest matching degree is selected as the user's corresponding constitution type. According to the user's constitution type, a combination of Chinese and Western medicine plan suitable for the user is selected. Combined with the intervention plan generated by the intervention plan recommendation module, a personalized intervention plan is generated, and the personalized intervention score of each intervention plan is calculated. If the user feedback is that the intervention effect is not good, the intervention plan recommendation module is used to optimize the plan and update the intervention strategy.

[0079] In this embodiment, if the user has a Yin deficiency constitution, high FSH and low E2, soy isoflavones combined with acupuncture are recommended to regulate endocrine;

[0080] If the user has a yang deficiency constitution and low LH and low BMD, it is recommended to take yang-warming Chinese medicine combined with calcium supplements;

[0081] If the user has a Qi deficiency constitution and low SI, Qi-boosting Chinese medicine combined with yoga training is recommended;

[0082] If the user has a phlegm-damp constitution and has a high BMI and high blood lipids, it is recommended to use Chinese medicine to clear heat and dampness combined with a low-carb diet.

[0083] The health community mutual aid module is used to receive user health data actively uploaded by users and establish a mutual aid community for patients; the exercise guidance module guides users' exercise in real time through bone key point detection.

[0084] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A menopausal syndrome health management platform based on big data, characterized in that: include: Marker monitoring module, data collection and processing module, health portrait modeling module, personal health assessment module, menopause prediction module, intervention plan recommendation module, medical data protection module, digital health interaction module, combined intervention module, health community mutual assistance module and exercise guidance module; The marker monitoring module uses home testing equipment to mine menopause-specific biomarkers; The data acquisition and processing module integrates and processes physiological indicators, wearable device data, behavioral data, environmental data, psychological assessment and genetic information to construct a menopause-specific data set; The health portrait modeling module dynamically builds a personalized user health portrait based on the menopause-specific data set; The personal health assessment module combines the user's real-time data to assess the severity of the user's menopausal symptoms; The menopause prediction module predicts the future menopause symptoms of the user and estimates their severity based on the user's health profile and the evaluation results of menopause symptoms; The intervention plan recommendation module dynamically optimizes the intervention plan based on user data and feedback; The medical data protection module is used to encrypt the data collected by the data collection and processing module and locally update the parameters of the menopause prediction module; The digital health interaction module is used to identify the user's emotional state and dynamically adjust the communication strategy; The combined intervention module is used to associate traditional Chinese medicine theory with Western medicine biological indicators and recommend personalized integrated Chinese and Western medicine programs; The health community mutual aid module is used to receive user health data actively uploaded by users and establish a patient mutual aid community; The movement guidance module guides the user's movement in real time through skeleton key point detection.

2. A menopausal syndrome health management platform based on big data according to claim 1, characterized in that: The specific steps of the health portrait modeling module to dynamically construct a personalized user health portrait are as follows: S1.1: Arrange the data in each group of the menopause-specific dataset in chronological order, extract the characteristic information in each group of data through principal component analysis, standardize the characteristic information of different dimensions, calculate the characteristic contribution of each characteristic information, and integrate the processed characteristic information of each group to construct a feature set; S1.2: Split the feature set into multiple groups of time series slices according to a preset fixed-length window, then add position codes to each time series slice, calculate the query, key, and value matrices of each group of time series slices, and independently calculate the query, key, and value matrices on multiple attention heads; S1.3: Calculate the attention scores of each group of time series slices through the query matrix and the key matrix, and establish the attention score matrix. Then use the Softmax function to normalize the attention scores, and perform weighted summation on the processed attention scores of each group. The weighted summation result of each attention head is used as the final attention output. S1.4: The final attention results output by each attention head are normalized through LayerNorm, and then the residuals are combined. After that, the combined attention values ​​are nonlinearly transformed through a fully connected neural network, and the Transformer-encoded time series feature matrix is ​​output. Each feature data in the time series feature matrix is ​​input into the LSTM model. S1.5: The LSTM model processes each feature data through a gating mechanism, and outputs the health status prediction value of the corresponding user through the fully connected layer of the model. It calculates the Euclidean distance between data points at different times and uses it as the high-dimensional distance between each data point. It also uses Gaussian distribution to perform high-dimensional distribution modeling on the similarity between each data point, and uses Cauchy distribution to perform low-dimensional distribution modeling on each data point. S1.6: The difference between the high-dimensional distribution and the low-dimensional distribution is calculated through the KL divergence function. Based on the difference values, the low-dimensional coordinates are updated through the gradient descent algorithm, and the corresponding 2D or 3D low-dimensional representation is generated, including different stages of menopausal health status, visualized health trends, and personalized health recommendations.

3. A menopausal syndrome health management platform based on big data according to claim 2, characterized in that: The specific steps of the personal health assessment module to assess the severity of the user's menopausal symptoms are as follows: S2.1: Collect medical knowledge data from various data sources including medical literature, medical guidelines, electronic health records, medical terminology, and clinical scoring systems, use NLP technology to convert unstructured medical literature into structured data, traverse each set of processed medical knowledge data, merge duplicate entries of the same disease, symptom, and treatment method in the data, and then map medical knowledge data from different sources into corresponding standard medical terms; S2.2: Extract the entities of diseases, symptoms, physiological indicators, treatment plans, lifestyle factors and environmental factors from the processed medical knowledge data through NRE technology, identify the relationship between the entities, establish knowledge triples in the form of head entity, relationship and tail entity, and store the triple data in GraphDB to establish the corresponding medical knowledge graph library; S2.3: Collect information on menopausal symptoms in the medical knowledge graph and calculate the user's clinical score , if 0≤ ≤6, the user is judged to have no menopausal symptoms. ≤15, the user is judged to have mild menopausal symptoms. ≤30, the user is judged to have moderate menopausal symptoms. , the user is judged to have severe menopausal symptoms, among which the clinical score The specific calculation formula is as follows: ； In the formula, represents the menopausal symptom score, Represents the total number of symptoms, Representative The weight of the symptoms, Representative The severity of the symptoms; S2.4: Based on the real-time collected physiological data, behavioral data, environmental data, and psychological state data of each user, the real-time data is mapped to the symptom severity correction factor through linear regression. Then, based on the symptom severity correction factor, the score of each menopausal symptom is corrected. Then, the similarity between symptoms is calculated based on the medical knowledge graph library, and the potential association between symptoms is inferred through logical rules. Then, the clinical score is calculated through K-Means. Perform final classification and output the severity of the user's final menopausal symptoms.

4. A menopausal syndrome health management platform based on big data according to claim 3, characterized in that: The specific steps of the intervention plan recommendation module to dynamically optimize the intervention plan are as follows: S3.1: Based on the multimodal data of users collected by the data acquisition and processing module, the corresponding health state space is constructed, and then the corresponding action space is established according to the intervention measures of adjusting hormone replacement therapy, recommending specific dietary adjustments, suggesting increasing aerobic exercise, recommending psychological intervention, and environmental optimization. The intervention plan is composed of different intervention measures. The reward function is established according to the symptom score, mental health score, user compliance change, and intervention cost to reflect the impact of the intervention plan on the health state; S3.2: Decompose the intervention plan taken by the current user into an action set consisting of different intervention measures, calculate the intervention effect of the current intervention plan on the user's menopausal symptoms through the reward function, and then calculate the probability distribution of each intervention measure in the action space based on the current user's health status; S3.3: At different intervention stages, the strategy network extracts multiple groups of intervention measures through the roulette wheel selection method based on the probability distribution of each intervention measure, and establishes corresponding intervention plans. Then, the intervention plans are executed, the changes in the health status of the user after the intervention are simulated and recorded, the intervention effect of the selected intervention plan is calculated through the reward function, and the intervention plans and their intervention effects constructed through multiple rounds of selection are recorded; S3.4: The value network calculates the cumulative discounted rewards and temporal differences of each intervention plan at different stages, and calculates the advantage estimate of each intervention plan through the cumulative discounted rewards and temporal differences, and then calculates the probability change of the new intervention plan for the current action relative to the old intervention plan, and calculates the total loss value of the strategy network and the value network based on the advantage estimate and probability change; S3.5: The Adam optimizer is used to optimize the parameters of the policy network and the value network. Then, strategy selection, advantage estimation, and parameter update are continuously performed until the change in the total loss value converges to the preset range. The iteration is stopped and the final generated intervention plan is output for real user intervention recommendations. User feedback is monitored in real time, and new data is recorded for subsequent intervention plan updates.

5. A menopausal syndrome health management platform based on big data according to claim 4, characterized in that: The specific steps of the medical data protection module locally updating the parameters of the menopause prediction module are as follows: S4.1: The medical data protection module initializes the global model parameters on the central server. At the same time, each medical institution establishes its local medical data set and local model respectively. The local medical data sets are not shared, and Paillier homomorphic encryption is used for local data. S4.2: The central server sends the current global model parameters to each medical institution. The local model of each medical institution uses the current global model parameters as its own initial parameters and divides the local medical data set into a training set and a test set. S4.3: Input the training set into the local model, the local model forward propagates the data in the training set, and outputs the prediction results of the user's menopausal symptoms. The loss value between the prediction result and the actual symptoms is calculated by the mean square error function. Then, based on the chain rule, the loss value is back-propagated from the output layer of the local model, and the gradient value of the loss value for the parameters of each network layer of the local model is calculated. At the same time, homomorphic encryption is used to protect the calculated gradient information; S4.4: Each medical institution uploads the encrypted gradient information to the central server, which then sums the gradients of each gradient information in an encrypted state to obtain the corresponding global gradient. The central server decrypts the calculated global gradient and updates the global model parameters through the gradient descent algorithm, and then encrypts the updated global model parameters. S4.5: The encrypted global model parameters are sent back to each medical institution. Each medical institution decrypts the model parameters and continues a new round of local training. The parameter sending, gradient calculation and global model parameter update are repeated until the loss value of the global model is lower than the preset threshold. The final global model parameters are used as the optimal parameters and sent to each medical institution. After that, the local model of each medical institution updates its own parameter information based on the optimal parameters and predicts the user's menopausal symptoms based on real-time user data.

6. A menopausal syndrome health management platform based on big data according to claim 1, characterized in that: The specific steps of recommending a personalized integrated Chinese and Western medicine program in the combined intervention module are as follows: S5.1: Collect the user's physical characteristics through the TCM physical constitution questionnaire and calculate the user's physical constitution score ,like If the user is within the preset physical condition threshold range, it indicates that the user has the physical condition type corresponding to the threshold, where the physical condition types include Yin deficiency, Yang deficiency, Qi deficiency, blood stasis and phlegm dampness; S5.2: Extract the user's serum hormone level, cortisol level, bone density and metabolic parameters, and calculate the hormone imbalance index and stress index ,like If the hormone imbalance is higher than the preset threshold, it indicates that the user has a high degree of hormone imbalance. If the value is higher than the preset upper threshold, it indicates that the user is in a state of severe stress. If it is lower than the preset lower threshold, it indicates that the user has chronic fatigue syndrome; S5.3: Use the expert knowledge base to establish a constitution-physiological index association table, and calculate the matching degree of different Chinese and Western medicine characteristics by weighted calculation, then select the constitution category with the highest matching degree as the constitution type corresponding to the user, and select the Chinese and Western medicine combination plan suitable for the user according to the user's constitution type, and combine it with the intervention plan generated by the intervention plan recommendation module to generate a personalized intervention plan; S5.4: Calculate the personalized intervention score for each intervention plan. If the user feedback shows that the intervention effect is not good, optimize the plan through the intervention plan recommendation module and update the intervention strategy.

Images