Cardiovascular metabolism health prediction platform based on big data

Abstract

The invention discloses a cardiovascular metabolism health prediction platform based on big data, and aims to integrate multi-source heterogeneous data and realize accurate prediction of cardiovascular metabolism health risks by constructing a hybrid prediction model combining a static feature sub-network and a dynamic feature sub-network. The static feature sub-network is used for processing data such as gene information, age and gender which do not change along with time, the dynamic feature sub-network is used for processing time sequence data such as heart rate and blood pressure, the two sub-networks generate joint features through weighted fusion, and the joint features are input into the classification network to output risk probability. The platform adopts a random forest algorithm to perform feature selection, the correlation and prediction efficiency of model input features are improved, and user privacy protection is realized through a federated learning framework. The platform provides dynamic trend analysis of health risk scores, outputs personalized health intervention suggestions including lifestyle adjustment and medication guidance, and generates real-time early warning for abnormal fluctuation of key indexes.

Description

Technical Field

[0001] The present invention relates to the technical field of health prediction platforms, and in particular to a cardiovascular metabolic health prediction platform based on big data. Background Art

[0002] Cardiovascular disease (CVD) is one of the leading causes of death and disability worldwide. Its main risk factors include hypertension, diabetes, obesity, unhealthy lifestyles, and genetic factors. With the aging of society and changes in lifestyles, cardiovascular and metabolic health issues are becoming more complex and diverse. Early screening and accurate prediction of diseases have become a research hotspot in the current medical and health field.

[0003] However, the occurrence of cardiovascular metabolic diseases is often the result of a combination of factors, including genetic information, daily health data (such as heart rate, blood pressure) and environmental influences. Traditional diagnostic methods mainly rely on the doctor's experience and static biochemical indicators (such as blood lipids and blood sugar levels), which are difficult to fully reflect the patient's dynamic health status and cannot predict disease risks in a timely manner.

[0004] (1) Insufficient data collection and integration

[0005] Most existing cardiovascular health management technologies rely on a single data source (such as medical records or wearable device data) and cannot comprehensively assess health status by integrating multi-source data (such as genome, environmental data, lifestyle habits, etc.). The integration and analysis of multi-source heterogeneous data has the following problems:

[0006] Data formats and sources are heterogeneous and lack unified standards;

[0007] Dynamic data (such as heart rate changes) and static data (such as genetic information) cannot be effectively combined, resulting in incomplete information in the prediction model.

[0008] (2) Insufficient fusion of dynamic and static features

[0009] Traditional prediction models are often based only on static data (such as the patient's age and gender), ignoring the dynamic characteristics of time series data such as heart rate and blood pressure. However, the changing trends of dynamic data can often reveal early signs of disease development, but existing models are unable to fully utilize these features, resulting in insufficient prediction accuracy.

[0010] (3) Conflict between model training and privacy protection

[0011] In practical applications, user privacy protection has become an important challenge for health data applications. Traditional centralized model training requires the collection of users' original data, which may lead to the risk of user data privacy leakage. Existing privacy protection measures often reduce data accuracy at the expense of model prediction results.

[0012] (4) Insufficient visualization and personalized intervention

[0013] The existing health management platform lacks intuitiveness in presenting data analysis results and cannot clearly show users the dynamic changes in their health status. At the same time, the health intervention suggestions for users are too general and lack personalized lifestyle guidance and real-time early warning mechanisms.

[0014] To this end, we urgently need to design a cardiovascular metabolic health prediction platform based on big data to solve the above problems. Summary of the invention

[0015] The purpose of the present invention is to provide a cardiovascular metabolic health prediction platform based on big data to address the deficiencies of the prior art and to solve the problems raised in the background technology.

[0016] To achieve the above objectives, the present invention provides the following technical solutions: a cardiovascular metabolic health prediction platform based on big data, comprising the following modules:

[0017] Data collection and integration module: used to collect and integrate multi-source heterogeneous data, including electronic health records, wearable device data, genomic data, lifestyle questionnaire data and environmental data;

[0018] Data analysis and storage module: used to clean, standardize and store the multi-source data, and use a distributed computing framework to implement data preprocessing and feature extraction;

[0019] Prediction algorithm and model module: used to build a dynamic cardiovascular metabolic health risk prediction model based on time series, and achieve high-precision prediction through multi-level feature learning; the prediction algorithm and model module uses a combination of static feature sub-network and dynamic feature sub-network to build a hybrid prediction model, the model includes: Static feature sub-network: used to process static input data, using a fully connected network to extract features, the mathematical formula is:

[0020] Z s =o(W s ·X s +b s )

[0021] Where: Z s : Extraction results of static features;

[0022] X s :Static input feature vector, including genetic information, gender and age;

[0023] W s : Weight matrix, used to connect the input layer and the hidden layer;

[0024] bs : bias vector;

[0025] σ: activation function;

[0026] Dynamic feature subnetwork: used to process time series data, using long short-term memory network to extract dynamic features, the formula is:

[0027] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0028] C t =f t ⊙C t-1 +i t ⊙tanh(W C ·[h t-1 ,x t ]+b C )

[0029] h t =o t ⊙tanh(C t )

[0030] Where: x t : Input features at time t; h t-1 : The hidden state at time t-1; f t 、i t , o t : are the activation values ​​of the forget gate, input gate and output gate respectively; C t : the unit state at time t; W f , W C : The corresponding weight matrix; b f , b C : corresponding bias vector; ⊙: element-wise multiplication; tanh: hyperbolic tangent activation function; σ: sigmoid activation function;

[0031] Visualization and user interaction module: used to provide users with personalized health risk prediction reports and intervention suggestions;

[0032] Health intervention and feedback module: used to collect user feedback data and optimize the prediction model based on online learning.

[0033] As a preferred technical solution of the present invention, the static feature Z s and dynamic characteristics h t Generate joint feature vector Z through weighted fusion c , the formula is:

[0034] Zc =α·Z s +(1-α)·h t

[0035] Where: Z c : Joint feature vector, representing the comprehensive representation of static features and dynamic features;

[0036] α: static feature weight parameter, ranging from 0≤α≤1, dynamically adjusted through model training;

[0037] h t : Dynamic characteristics at time t.

[0038] As a preferred technical solution of the present invention, the joint feature Z c It is input into the classification network to predict cardiovascular metabolic health risks. The output formula of the classification network is:

[0039]

[0040] in: The predicted disease risk probability ranges from [0,1]; W c : weight matrix of classification network;

[0041] b c : Bias of classification network; σ: Sigmoid activation function.

[0042] As a preferred technical solution of the present invention, the platform optimizes the model through the following cross entropy loss function:

[0043]

[0044] Where: L: loss value of the model; N: number of training samples; y i : The actual label of the i-th sample (0 or 1); The predicted probability of the i-th sample.

[0045] As a preferred technical solution of the present invention, in order to improve the optimization efficiency of the cross entropy loss function, the input feature set is screened by the following method based on feature importance analysis to reduce redundant features and reduce model complexity: Calculate the importance score of each feature:

[0046] Where: Gini: Gini impurity; p k : The proportion of samples in the kth category; K: The total number of categories; All input features are sorted according to the feature importance scores, and the top M features in importance are selected for the input of the static feature sub-network and the dynamic feature sub-network.

[0047] As a preferred technical solution of the present invention, the feature importance analysis is implemented based on the random forest algorithm, which specifically includes the following steps:

[0048] Build a random forest model and calculate the cumulative importance of each feature at the split node;

[0049] Sort by feature importance and select the top M features to input into the static feature sub-network and the dynamic feature sub-network.

[0050] As a preferred technical solution of the present invention, the health intervention and feedback module optimizes model parameters through online learning, and the specific algorithm is:

[0051]

[0052] Where: θ t : Current model parameters; η: learning rate; The gradient of the current loss function.

[0053] As a preferred technical solution of the present invention, the platform protects user privacy through a federated learning framework. The specific steps include: training the model locally to generate parameters θ k ; Use encryption algorithm to upload encrypted θ k ; Model parameters for aggregating users on the server side:

[0054]

[0055] Where: K: total number of users; θ k : Model parameters for the kth user.

[0056] As a preferred technical solution of the present invention, the visualization and user interaction module includes:

[0057] Provide dynamic trends of health risk scores;

[0058] Output personalized health intervention recommendations, including lifestyle adjustment recommendations and medication guidance;

[0059] Generate real-time early warning alerts for abnormal fluctuations in key indicators.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The present invention effectively solves the problem of insufficient single data source in the prior art by collecting and integrating multi-source heterogeneous data (such as electronic health records, wearable device data, genome data, lifestyle questionnaire data and environmental data). The platform adopts standardized processing and distributed storage technology, which can efficiently integrate static and dynamic data and provide comprehensive basic data support for disease risk prediction.

[0062] The hybrid prediction model designed by the present invention combines static feature subnetworks with dynamic feature subnetworks, and can simultaneously process static features such as genetic information, age, and gender, as well as time series features such as heart rate and blood pressure, fully exploring the changing patterns of dynamic features and the influence of static features. Through feature weighted fusion and deep learning algorithms, the accuracy of cardiovascular metabolic health risk prediction is significantly improved, providing users with more reliable health assessments.

[0063] The present invention adopts a federated learning framework to protect user data privacy while achieving global optimization of the model, avoiding the risk of uploading original data. In addition, the platform provides dynamic trend analysis of health risk scores, personalized intervention suggestions and real-time early warning functions, helping users to intuitively understand their own health status and take targeted measures in a timely manner, with good user experience and practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a system block diagram of a cardiovascular metabolic health prediction platform based on big data proposed by the present invention. DETAILED DESCRIPTION

[0065] 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.

[0066] The following is combined with Figure 1 With reference to the accompanying drawings and a number of embodiments, specific embodiments of the present invention are described in detail.

[0067] Example 1: Architecture design of cardiovascular metabolic health prediction platform based on big data

[0068] The cardiovascular metabolic health prediction platform of the present invention is composed of the following modules, and the specific implementation of each module is as follows:

[0069] Data collection and integration module:

[0070] Data source: Electronic health records (EHR): including patient medical history, physical examination reports (such as blood sugar, blood lipids), examination data, etc.

[0071] Wearable device data: collect real-time dynamic data such as heart rate, blood pressure, and blood oxygen through smart watches, blood pressure monitors and other devices.

[0072] Genomic data: including whole genome sequencing or SNP testing data, used to assess genetic risk.

[0073] Lifestyle questionnaire: Users submit it through the platform, including diet, exercise, smoking, drinking, sleep quality, etc.

[0074] Environmental data: Obtain environmental information such as air quality (PM2.5), temperature and humidity.

[0075] Data cleaning and integration:

[0076] Perform missing value completion and outlier detection on the collected data (such as filtering abnormally high heart rate values);

[0077] Data standardization (such as standardizing blood pressure units to mmHg);

[0078] One-Hot encoding is performed on the genetic data and stored in a distributed database.

[0079] Data analysis and storage module:

[0080] Data storage uses a distributed storage framework (such as HDFS) to support fast access and processing of multi-source data.

[0081] Data analysis includes:

[0082] Feature extraction: extract mean, variance, extreme value, and frequency domain features from heart rate time series;

[0083] Feature dimensionality reduction: Use the random forest algorithm to calculate feature importance and remove features with low relevance.

[0084] Prediction algorithm and model module:

[0085] Static feature subnetwork: used to process static features such as genetic data, age, gender, etc., using a fully connected network to extract features:

[0086] Z s =σ(W s ·X s +b s )

[0087] Among them, X s is the static input feature, W s is the weight matrix, b s is the bias and σ is the activation function.

[0088] Dynamic feature subnetwork: used to process time series data (such as daily heart rate and blood pressure), using the LSTM model to extract dynamic features. The formula is as follows:

[0089] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0090] C t =f t ⊙C t-1 +i t ⊙tanh(W C ·[h t-1 ,x t ]+b C )

[0091] h t =o t ⊙tanh(C t )

[0092] Where: x t : Input features at time t; h t-1 : The hidden state at time t-1; f t 、i t , o t : are the activation values ​​of the forget gate, input gate and output gate respectively; C t : the unit state at time t; W f , W C : The corresponding weight matrix; b f , b C : corresponding bias vector; ⊙: element-wise multiplication; tanh: hyperbolic tangent activation function; σ: sigmoid activation function.

[0093] Joint feature fusion and classification network: static feature Z s and dynamic characteristics h t Weighted Fusion:

[0094] Z c =α·Z s +(1-α)·h t

[0095] Where: Z c : Joint feature vector, representing the comprehensive representation of static features and dynamic features;

[0096] α: static feature weight parameter, ranging from 0≤α≤1, dynamically adjusted through model training; h t : Dynamic characteristics at time t.

[0097] Joint feature Z c Input classification network to generate prediction results

[0098]

[0099] in: The predicted disease risk probability ranges from [0,1]; Wc : weight matrix of classification network;

[0100] b c : Bias of classification network; σ: Sigmoid activation function.

[0101] Visualization and user interaction module:

[0102] Provide dynamic trend charts of health risk scores to help users track risk changes;

[0103] Output personalized intervention recommendations, including lifestyle adjustments and medication guidance;

[0104] Generate real-time warnings, such as push reminders when blood pressure or heart rate fluctuates abnormally.

[0105] Health intervention and feedback module: After the user executes the intervention plan, he / she submits feedback data (such as blood pressure changes and step count records) through the platform, and the platform uses this data to dynamically update the model.

[0106] Federated learning and privacy protection module: Users complete model training on local devices and generate local parameters θ k ; The local parameters are encrypted and uploaded to the server, and the server aggregates them to obtain the global parameters:

[0107]

[0108] Where: K: total number of users; θ k : Model parameters for the kth user.

[0109] Example 2: Disease prediction based on the combination of genes and dynamic features

[0110] Application scenario: A patient uploads his health data, including: static data: age 50, gender male, gene test results show a positive gene mutation; dynamic data: blood pressure records for the past month (systolic pressure 135mmHg, diastolic pressure 85mmHg); life habits: salty diet, average daily steps 4000 steps.

[0111] Implementation steps: Feature extraction:

[0112] The platform extracts dynamic features (such as blood pressure trends) and static features (such as gene coding, age, and gender); through feature importance analysis, it selects the 50 features that have the greatest impact on prediction. Risk prediction: static features are input into the fully connected network to generate Z s , dynamic feature input LSTM network generates h t ; Static and dynamic features are fused to generate Z c , and input the classification network to generate prediction results Intervention recommendations: Lifestyle adjustments: Reduce salt intake and walk 8,000 steps per day; users are advised to monitor blood pressure and consult a doctor about the use of antihypertensive medications.

[0113] Example 3: Personalized health intervention and feedback mechanism

[0114] Application background: A patient received a 3-month health intervention through the platform, and the initial cardiovascular metabolic health risk was

[0115] Implementation process: Intervention plan:

[0116] The platform recommends that patients walk 6,000 steps a day, reduce high-sodium diets, and monitor blood pressure changes; the platform generates health reports every week to track patients' improvements.

[0117] Feedback and model update: Patients uploaded their feedback data (such as step count and blood pressure changes). After 3 months, systolic blood pressure dropped by 10 mmHg and the average number of steps per day reached 7,500.

[0118] The platform uses feedback data to update model parameters through online learning algorithms to improve prediction accuracy.

[0119] Example 4: Implementation of Federated Learning and Privacy Protection

[0120] Application background: To protect user privacy, the platform adopts the federated learning method.

[0121] Implementation process: Local model training:

[0122] Each user’s device trains the model independently to generate local parameters θ k .

[0123] The data is not uploaded, only the encrypted model parameters are uploaded.

[0124] Server-side aggregation: The server receives local parameters from K users and aggregates them to obtain a global model:

[0125]

[0126] Model delivery: The aggregated global model is delivered to the user device to update the local model and achieve personalized prediction.

[0127] Example 5: Dynamic trend visualization and real-time warning

[0128] Application scenario: A patient uploads real-time heart rate and blood pressure data through a wearable device.

[0129] Functions implemented: Dynamic trend analysis: The platform draws a time-dynamic curve of the health risk score based on the patient's data over the past 6 months to show the trend of risk changes.

[0130] Real-time warning: If the patient's heart rate exceeds 100 beats / minute for 10 consecutive minutes, or the systolic blood pressure exceeds 140 mmHg, the platform will push a warning message and recommend stopping strenuous exercise.

[0131] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in the field. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A cardiovascular metabolic health prediction platform based on big data, characterized by: Includes the following modules: Data collection and integration module: used to collect and integrate multi-source heterogeneous data, including electronic health records, wearable device data, genomic data, lifestyle questionnaire data and environmental data; Data analysis and storage module: used to clean, standardize and store the multi-source data, and use a distributed computing framework to implement data preprocessing and feature extraction; Prediction algorithm and model module: used to build a dynamic cardiovascular metabolic health risk prediction model based on time series, and achieve high-precision prediction through multi-level feature learning; the prediction algorithm and model module uses a combination of static feature sub-network and dynamic feature sub-network to build a hybrid prediction model, the model includes: Static feature sub-network: used to process static input data, using a fully connected network to extract features, the mathematical formula is: Z s =σ(W s ·X s +b s ) Where: Z s : Extraction results of static features; X s :Static input feature vector, including genetic information, gender and age; W s : Weight matrix, used to connect the input layer and the hidden layer; b s : bias vector; σ: activation function; Dynamic feature subnetwork: used to process time series data, using long short-term memory network to extract dynamic features, the formula is: f t =σ(W f ·[h t-1 ,x t ]+b f ) C t =f t ⊙C t-1 +i t ⊙tanh(W C ·[h t-1 ,x t ]+b C ) h t =o t ⊙tanh(C t ) Where: x t : Input features at time t; h t-1 : The hidden state at time t-1; f t 、i t , o t : are the activation values ​​of the forget gate, input gate and output gate respectively; C t : the unit state at time t; W f , W C : The corresponding weight matrix; b f , b C : corresponding bias vector; ⊙: element-by-element multiplication; tanh: hyperbolic tangent activation function; v: sigmoid activation function; Visualization and user interaction module: used to provide users with personalized health risk prediction reports and intervention suggestions; Health intervention and feedback module: used to collect user feedback data and optimize the prediction model based on online learning.

2. The cardiovascular metabolic health prediction platform based on big data as described in claim 1, characterized in that: The static characteristic Z s and dynamic characteristics h t Generate joint feature vector Z through weighted fusion c , the formula is: WITH c =α Z s +(1-α) h t Where: Z c : Joint feature vector, representing the comprehensive representation of static features and dynamic features; α: static feature weight parameter, ranging from 0≤α≤1, dynamically adjusted through model training; h t : Dynamic characteristics at time t.

3. The cardiovascular metabolic health prediction platform based on big data according to claim 2, characterized in that: The joint feature Z c It is input into the classification network to predict cardiovascular metabolic health risks. The output formula of the classification network is: in: The predicted disease risk probability ranges from [0,1]; W c : weight matrix of classification network; b c : Bias of classification network; σ: Sigmoid activation function.

4. The cardiovascular metabolic health prediction platform based on big data according to claim 3, characterized in that: The platform optimizes the model using the following cross entropy loss function: Where: L: loss value of the model; N: number of training samples; y i : The actual label of the i-th sample (0 or 1); The predicted probability of the i-th sample.

5. The cardiovascular metabolic health prediction platform based on big data according to claim 4, characterized in that: In order to improve the optimization efficiency of the cross entropy loss function, the input feature set is screened by the following method based on feature importance analysis to reduce redundant features and reduce model complexity: Calculate the importance score of each feature: Where: Gini: Gini impurity; p k : The proportion of samples in the kth category; K: The total number of categories; All input features are sorted according to the feature importance scores, and the top M features in importance are selected for the input of the static feature sub-network and the dynamic feature sub-network.

6. The cardiovascular metabolic health prediction platform based on big data according to claim 6, characterized in that: The feature importance analysis is implemented based on the random forest algorithm, which specifically includes the following steps: Build a random forest model and calculate the cumulative importance of each feature at the split node; Sort by feature importance and select the top M features to input into the static feature sub-network and the dynamic feature sub-network.

7. The cardiovascular metabolic health prediction platform based on big data according to claim 5, characterized in that: The health intervention and feedback module optimizes model parameters through online learning. The specific algorithm is: Where: θ t : Current model parameters; η: learning rate; The gradient of the current loss function.

8. The cardiovascular metabolic health prediction platform based on big data according to claim 7, characterized in that: The platform protects user privacy through a federated learning framework. The specific steps include: training the model locally to generate parameters θ k ; Use encryption algorithm to upload encrypted θ k ; Model parameters for aggregating users on the server side: Where: K: total number of users; θ k : Model parameters for the kth user.

9. The cardiovascular metabolic health prediction platform based on big data according to claim 1, characterized in that: The visualization and user interaction module includes: Provide dynamic trends of health risk scores; Output personalized health intervention recommendations, including lifestyle adjustment recommendations and medication guidance; Generate real-time early warning alerts for abnormal fluctuations in key indicators.

Images