A health information pushing and physical examination optimization method

Abstract

This invention belongs to the field of health information processing and is a method for health information push and physical examination optimization, including the following steps: S1: The user completes physical examination registration and basic information input, performs physical examination registration operations, and supplements static basic information and health requests; S2: Generate personalized physical examination adaptation parameters, construct user profiles, allocate priority weights for physical examination items, and verify their rationality; S3: Smart terminal acquires and preprocesses multi-dimensional data, acquires multi-source data, and performs standardized processing; S4: Large model dynamically integrates and analyzes to output optimization results, fuses and models the standardized data, iteratively optimizes the prediction model, and outputs analysis results hierarchically; S5: Real-time adjustment and health warning during the physical examination, adjusts the order of physical examinations according to the analysis results, links devices to monitor vital signs, and responds to user feedback; S6: Physical examination result comparison, correction, and feedback iteration, collects the current physical examination results and compares and corrects them with the analysis results, and updates the user profile and model parameters.

Description

Technical Field

[0001] This invention belongs to the field of health information processing, and in particular to a method for health information push and physical examination optimization. Background Technology

[0002] The current health checkup market faces problems such as diversified user needs, insufficient accurate early warning of health risks, and a disconnect between health checkups and subsequent health management. There is an urgent need to build an intelligent and personalized service system covering the entire process to meet the differentiated health needs of different groups and realize the proactive health concept of preventing disease.

[0003] Current technologies in health information processing and physical examination optimization have gradually integrated AI and big data capabilities, forming a multi-dimensional technology application landscape. Online medical platforms primarily rely on algorithms such as collaborative filtering and content recommendation to build user profiles by collecting data such as user registration information, browsing behavior, and health questionnaires, enabling personalized health information delivery. In the physical examination field, AI-powered intelligent body solutions have emerged. These solutions can customize physical examination packages before the examination based on factors such as age, gender, and past medical history; improve process efficiency during the examination through intelligent triage and group booking scheduling optimization; and generate individualized reports and provide health management suggestions after the examination using large-scale models. Some technologies have also achieved functions such as image analysis and simultaneous multi-organ risk assessment, initially constructing a closed loop of health management through data fusion. However, the core focus remains on optimizing single aspects, and a fully adaptive system with real-time linkage of multi-source dynamic data has not yet been formed.

[0004] Therefore, existing health checkup technologies mostly adopt fixed procedures and standardized information delivery models. Data collection is limited to historical health checkup records and basic physiological data, lacking a comprehensive consideration of users' dynamic health status, personalized needs, and the resource status of health checkup centers. Furthermore, the fixed procedures during the health checkup process and the lack of real-time adjustment mechanisms create information silos between health information delivery and checkup result feedback, resulting in a poor user experience. Summary of the Invention

[0005] This invention proposes a method for health information push and physical examination optimization.

[0006] A method for health information delivery and physical examination optimization includes the following steps: S1: The user completes the physical examination registration and basic information entry, performs the physical examination registration operation, and supplements static basic information and health requirements; S2: Generate personalized physical examination adaptation parameters, construct a user profile based on the basic information and health needs, assign priority weights to physical examination items and verify their rationality; S3: Smart terminal multi-dimensional data acquisition and preprocessing, acquire multi-source data according to the adaptation parameters and perform standardized processing to obtain standardized data; S4: The large model performs dynamic comprehensive analysis and outputs optimization results. It integrates and models the standardized data obtained in S3 and iteratively optimizes the prediction model, outputting analysis results and health information related to the current physical examination item in a hierarchical manner. S5: Real-time adjustment and health warning during physical examination. The order of physical examination is dynamically adjusted according to the analysis results obtained in S4. The linked equipment monitors vital signs in real time and pushes the health information and responds to user feedback. S6: Physical examination result comparison, correction and feedback iteration: Collect the current physical examination results and compare and correct them with the analysis results, and update the user profile and model parameters.

[0007] Preferably, in step S1, a physical examination registration operation is performed and static basic information and health requests are supplemented. The static basic information includes the user's age, basic medical history, and occupation-related data. The health requests include the need for preparing for pregnancy, the need for chronic disease management, and the core purpose of the physical examination. The static basic information and health requests together constitute a basic information package.

[0008] Preferably, in step S2, a user profile is constructed based on the basic information and health needs. When constructing a multi-dimensional user profile, the following is included: the static data in the basic information package is integrated with the health needs to form a multi-dimensional user profile covering the user's health status, risk tendency, and demand orientation.

[0009] Preferably, in step S3, multi-source data is acquired according to the adaptation parameters and standardized. The multi-source data includes historical physical examination data obtained from the physical examination user historical database, physical examination recommendation order information obtained from the physical examination intelligent order recommendation module, and monitoring data collected from mobile health APP and sports APP. The monitoring data includes heart rate monitoring data, sleep time data, and exercise-related data.

[0010] Preferably, in step S3, multi-source data is acquired according to the adaptation parameters and standardized. The standardization process includes cleaning the multi-source data using a preset standardization algorithm, filtering redundant data and outliers, and converting unstructured data into structured labels or quantitative indicators.

[0011] Preferably, the multimodal data fusion modeling in S4 adopts an attention mechanism to fuse features of structured and unstructured data and explore potential correlations between data of different dimensions.

[0012] S41: Multimodal data fusion modeling adopts an attention mechanism to integrate the standardized structured data and unstructured data features from S3 to explore the potential correlations between data of different dimensions; S42: Based on the fused dataset in S41, the model parameters are updated by improving the incremental learning algorithm to perform real-time iterative optimization of the prediction model; S43: Based on the prediction model analysis results optimized from S42, the analysis results are classified according to the urgency of health risks and user awareness. High-risk items are given priority push permissions and are accompanied by a doctor intervention reminder icon, while routine items are accompanied by personalized optimization suggestions based on user profiles.

[0013] Preferably, the dynamic optimization of the physical examination order in S5 includes real-time collection of physical examination user's physical status data and physical examination center traffic data, and adjustment of the physical examination execution order based on the priority weight of the physical examination items. When the user is in a fatigued state, non-invasive rapid physical examination items are given priority.

[0014] Preferably, the health warning in S5 includes establishing a real-time communication link between the smart terminal and the physical examination equipment, synchronously collecting real-time vital sign data and comparing it with a preset threshold, triggering a graded warning mechanism, with different warning levels corresponding to different reminder methods and response measures.

[0015] Preferably, updating the user profile and model parameters in S6 includes incorporating abnormal items from the current physical examination results into the user's health risk label system, supplementing user data sensitivity parameters, and updating the parameters and weight factors of the prediction model based on the corrected deviation values.

[0016] Preferably, the personalized adaptation parameters are linked and optimized with the acquisition of multi-source data. Monitoring data corresponding to high-weight physical examination items are collected first, and the completeness and accuracy of the monitoring data are fed back to the adaptation parameter verification process, triggering secondary verification of the adaptation parameters or adjustment of the weight of the corresponding physical examination items.

[0017] The present invention has the following beneficial effects: This invention integrates users' static basic information, dynamic health monitoring data, and the status of physical examination resources to accurately allocate the priority of physical examination items and data collection strategies, thereby significantly improving the targeting of physical examination services and the efficiency of process operation, and adapting to the differentiated needs of different user groups. This invention effectively avoids health risks during the physical examination process and improves the user's physical examination experience by real-time linkage with physical examination equipment to monitor vital signs, dynamically adjust the order of physical examinations, and quickly respond to user feedback. This invention forms a closed-loop optimization of health information push and physical examination services through dynamic iterative optimization of a large model and comparison and correction of physical examination results, providing users with more accurate health analysis results and personalized suggestions to assist in subsequent health management. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of a health information push and physical examination optimization method according to the present invention. Figure 2 This is a schematic diagram of the system interface of Embodiment 4 of the health information push and physical examination optimization method of the present invention. Figure 1; Figure 3 This is a schematic diagram of the system interface of Embodiment 4 of the health information push and physical examination optimization method of the present invention. Figure 2 . Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly described below in conjunction with the examples.

[0020] This method primarily addresses two major technical issues: First, existing technologies cannot integrate multi-source dynamic data to achieve personalized adaptation throughout the entire physical examination process, resulting in insufficient targeting of examination items, low process efficiency, and difficulty in meeting the differentiated needs of different users, such as patients with chronic diseases, working professionals, and the elderly; Second, existing technologies lack real-time health monitoring and dynamic adjustment mechanisms during the physical examination process, making it impossible to respond promptly to changes in users' physical condition and temporary requests, and the accuracy of health information push is insufficient, making it difficult to effectively assist users in making physical examination decisions and subsequent health management.

[0021] like Figure 1 As shown, this invention proposes a method for health information push and physical examination optimization, which specifically includes the following steps: S1. The user completes the physical examination registration and basic information entry, performs the physical examination registration operation, and supplements static basic information and health requirements; S2. Generate personalized physical examination adaptation parameters, construct a user profile based on the basic information and health needs, assign priority weights to physical examination items and verify their rationality; S3. Smart terminal multi-dimensional data acquisition and preprocessing: acquire multi-source data according to the adaptation parameters and perform standardized processing; S4. The large-scale dynamic comprehensive analysis outputs optimization results, fuses and models the standardized data, iteratively optimizes the prediction model, and outputs analysis results in a hierarchical manner. S5. Real-time adjustment and health warning during physical examination: Adjust the order of physical examination according to the analysis results, link the equipment to monitor vital signs and respond to user feedback. S6. Physical examination result comparison, correction and feedback iteration: Collect the physical examination results and compare and correct them with the analysis results, and update the user profile and model parameters.

[0022] Furthermore, step S1 also includes S1-1 accurate collection of basic information, specifically: collecting static data such as user age, basic medical history, occupation, etc., and capturing the user's recent health needs in real time. The static data and health needs together constitute a basic information package, wherein the health needs include needs for preparing for pregnancy, needs for chronic disease management, and the core purpose of physical examination. Furthermore, step S2 also includes S2-1, which involves accurately constructing a multi-dimensional user profile. Specifically, this involves modeling by integrating static data from the basic information package with health needs to form a multi-dimensional user profile that covers the user's health status, risk propensity, and demand orientation. Step S2 further includes S2-2 dynamic weight allocation of physical examination item priorities, specifically: setting weight factors based on the multi-dimensional user profile, and calculating the priority weight of each physical examination item using a weight allocation formula, wherein the weight allocation formula is... ,in Let i be the priority weight of the i-th physical examination item. The matching coefficient between the i-th individual test item and the user profile (value range 0-1). Let be the correlation coefficient between the i-th individual test item and the abnormal data of the previous year (value range 0-1). Let be the matching coefficient between the i-th physical examination item and the resources of the physical examination center (value range 0-1). , Main weighting factor It is an interactive weight factor, and satisfies ; Step S2 also includes S2-3 real-time rationality verification of the adaptation parameters, specifically: extracting similar user adaptation cases from the historical database of physical examination users, removing extreme values ​​in the weight allocation results, and verifying the matching degree between the weight allocation results and the equipment and doctor resources of the physical examination center to form the final personalized adaptation parameters. The personalized adaptation parameters include the weight of the physical examination items and the priority of data collection. Furthermore, step S3 also includes S3-1 targeted data collection, specifically: according to the data collection priority in the personalized adaptation parameters, the first data of the previous year's physical examination is obtained from the historical database of physical examination users, the second data of the physical examination recommendation order information table is obtained from the intelligent physical examination order recommendation module, and the third data is collected targeted from the mobile health APP and sports APP. The third data includes heart rate monitoring data, sleep time data and various exercise data. Step S3 further includes S3-2 data standardization processing, specifically: using the Z-score standardization algorithm to clean the first data, the second data, and the third data, filtering redundant data and outliers, converting the sports APP behavior logs into standardized activity volume indicators, and converting the unstructured data of the health APP into structured user health attention tags, forming a standardized multi-source dataset. Furthermore, step S4 also includes S4-1 multimodal data fusion modeling, specifically: using an attention mechanism to fuse the features of structured and unstructured data in the standardized multi-source dataset, constructing a multimodal fusion model, and mining the potential correlations between data of different dimensions; Step S4 further includes S4-2 real-time iterative optimization of the prediction model, specifically: updating the model parameters based on the improved incremental learning algorithm, wherein the parameter update formula is as follows: ,in For the updated model parameters, These are the current model parameters. These are the parameters from the previous model round. Based on the learning rate, This is a data credibility weighting factor (with a value range of 0.8-1.0, dynamically assigned based on the credibility of the data source). Based on current parameters and real-time collected data The gradient of the loss function is calculated. Momentum factor (value range 0.05-0.15); Step S4 also includes S4-3 hierarchical result precision output, specifically: classifying the analysis results according to the urgency of health risks and user awareness, with high-risk items corresponding to priority push permissions and additional doctor intervention reminder labels, and routine items accompanied by personalized optimization suggestions based on user profiles, forming the optimized first result; Furthermore, step S5 also includes S5-1 dynamic optimization of the physical examination order, which specifically involves: collecting user physical status data and physical examination center traffic data in real time, and dynamically adjusting the physical examination execution order based on the weight of the physical examination items in the optimized first result. Non-invasive rapid physical examination items are given priority when the user is fatigued, and popular physical examination items are diverted according to traffic data. Step S5 also includes S5-2 health warning during the physical examination process, specifically: establishing a real-time communication link between the smart terminal and the physical examination equipment, synchronously collecting real-time vital sign data output by the physical examination equipment, comparing the real-time vital sign data with a preset threshold, triggering a graded warning mechanism, a yellow warning is reminded through a pop-up window on the smart terminal, and a red warning is directly synchronized to the doctor's terminal at the physical examination center; Step S5 also includes S5-3 Real-time User Feedback and Rapid Response, specifically: collecting user comfort feedback and urgent needs during the physical examination process through a smart terminal pop-up questionnaire, generating a process adjustment plan within 10 seconds based on the feedback information, synchronizing it to the physical examination center system, and updating the physical examination execution plan. Furthermore, step S6 also includes S6-1 physical examination result comparison and correction, specifically: collecting the fourth data of this physical examination result, comparing the fourth data with the optimized first result dimension by dimension, calculating the deviation value and correcting the output error of the analysis model; Step S6 also includes S6-2 user profile and model parameter iterative update, specifically: incorporating the outliers in the fourth data into the user health risk label system, supplementing the multi-dimensional user profile dimensions, updating the parameters and weight factors of the prediction model based on the corrected deviation value, and synchronously updating the historical database of physical examination users. Furthermore, the personalized adaptation parameters and multi-dimensional data acquisition form a linkage optimization. Specifically, the data collection priority in the personalized adaptation parameters specifies the collection order of each type of monitoring data. Monitoring data corresponding to high-weight physical examination items are collected first, and redundant data without relevant requirements are automatically filtered by the algorithm. At the same time, the completeness and accuracy of the monitoring data are fed back to the adaptation parameter verification stage. When monitoring data is missing, the weight of the corresponding physical examination item is reduced, and when monitoring data is abnormal, a secondary verification of the personalized adaptation parameters is triggered. Furthermore, the analysis results of the large model and the real-time adjustment of the physical examination form a closed-loop interaction. Specifically, high-risk items in the optimized first result trigger the in-depth adjustment of the physical examination detection, increasing the number of detection indicators for the corresponding items, while low-risk items shorten the detection time through process optimization. The execution data and detection results after the real-time adjustment of the physical examination are sent back to the large model as incremental learning samples to correct prediction bias and optimize the model feature weight allocation. Furthermore, the physical examination results comparison and correction are iteratively linked with the user profile update. Specifically, the deviation data after comparison and correction is used to supplement the user data sensitivity parameters, the abnormal items in this physical examination are used to update the user's long-term health risk label, and the updated multi-dimensional user profile is directly used for the next round of personalized adaptation parameter generation, adjusting the weight allocation of the physical examination items and the data collection scope for the next year. Furthermore, the smart terminal and the physical examination equipment achieve real-time communication. Specifically, after the physical examination equipment completes the test, it automatically uploads the result data to the smart terminal and adapts it to the large model input standard through a data format conversion algorithm, eliminating the need for manual data entry. The smart terminal pushes the user adaptation information in the personalized adaptation parameters to the physical examination equipment, automatically adjusting the equipment's test parameters for elderly users and presets a comfort mode for highly sensitive users. Furthermore, the resources of the physical examination center are collaboratively adapted to personalized needs. Specifically, equipment malfunctions and doctor scheduling information are fed back to the personalized adaptation parameter generation stage in real time, automatically replacing unperformable examination items as alternatives, and high-risk items are accurately allocated based on the doctor's expertise. The smart terminal counts the frequency of user examination needs, forming demand popularity data, which provides data support for increasing investment in physical examination center equipment and adjusting medical staff scheduling.

[0023] Example 2 The technical features that distinguish this embodiment from Embodiment 1 are as follows: A method for health information push and physical examination optimization includes the following: In step S1, the user performs the physical examination registration operation through the offline registration terminal or online registration platform of the physical examination center. During the registration process, the system automatically associates the user's identity identifier, which includes ID card number, mobile phone number or biometric information. The biometric information is obtained through the collection module of the registration terminal, which includes a fingerprint sensor and a facial recognition camera. The collection process follows the data encryption transmission standard to ensure information security. After registration, a pop-up window on the smart terminal displays a basic information supplementation interface, collecting static data such as the user's age, basic medical history, and occupation. The basic medical history is entered in a structured manner, providing preset medical history tags for users to select, and also supporting custom input of medical history information not covered. Real-time capture of the user's recent health needs is achieved through a combination of user-initiated information and intelligent system inference. The system generates a candidate health needs list by analyzing the user's historical physical examination records and recent health consultation records. Users can check and confirm or supplement and modify the list. The static data and health needs together constitute a basic information package, which is stored in JSON data format. The fields include user ID, static data set, health needs set, and registration timestamp. The health needs set is clearly divided into three sub-fields: pre-pregnancy needs, chronic disease management needs, and the core purpose of the physical examination, ensuring that the data structuring meets the requirements of subsequent modeling.

[0024] Step S2 initiates the personalized health check-up adaptation parameter generation process based on the basic information package. First, it executes S2-1, a multi-dimensional user profile construction process, integrating static data and health needs from the basic information package to create a model. A weighted fusion algorithm assigns differentiated weights to different dimensions of data, with age and basic medical history having higher weights than occupation and health needs. These weights are determined using the analytic hierarchy process (AHP). This constructs a multi-dimensional user profile encompassing the user's health status, risk propensity, and needs orientation. Specifically, the health status dimension is quantitatively assessed using age and basic medical history from static data, generating a health status score of 0-100 using a health risk scoring algorithm. The risk propensity dimension is determined through correlation analysis between basic medical history and health needs; for example, users with a history of diabetes and whose health needs include blood glucose management are marked as having a high metabolic risk propensity. The needs orientation dimension directly maps to the core needs within the health requirements, forming a user profile tag set. Each tag includes three attributes: tag name, weight value, and confidence level.

[0025] Subsequently, the dynamic weight allocation of the S2-2 physical examination item priority is performed. Weighting factors are set based on the multi-dimensional user profile, and the priority weight of each physical examination item is calculated using a weight allocation formula. The weight allocation formula is as follows: ,in Let i be the priority weight of the i-th physical examination item. Let be the matching coefficient between the i-th physical examination item and the user profile, with a value ranging from 0 to 1. It is obtained by calculating the cosine similarity between the physical examination item label and the user profile label set. The specific calculation method is as follows: ,in The weight value of the k-th tag in the user profile. Let k be the association strength of the k-th label of the i-th individual test item; Let be the correlation coefficient between the i-th physical examination item and the abnormal data of the previous year, with a value range of 0-1. It is obtained by analyzing the causal relationship between the abnormal physical examination items of the previous year and the current physical examination item through association rule mining algorithm. The mining process adopts the Apriori algorithm, with the minimum support set to 0.2 and the minimum confidence set to 0.7. Let be the matching coefficient between the i-th physical examination item and the resources of the physical examination center, with a value ranging from 0 to 1. It is obtained by weighting three indicators: availability of physical examination equipment, professional matching degree of doctors, and examination time. The calculation formula is as follows: ,in Let i be the real-time availability rate of the device corresponding to the i-th physical examination item. To ensure the match between the doctor's expertise and the physical examination items, This represents the normalized value of the standard testing duration for each physical examination item. , Main weighting factor It is an interactive weight factor, and satisfies .

[0026] After completing the weight calculation, the S2-3 adaptation parameter real-time rationality verification is performed. This involves connecting to the historical database of physical examination users and extracting similar user adaptation cases based on screening criteria, including age ±5 years, health status score difference ≤10 points, consistent risk tendency labels, and similar health needs. This forms a set of similar cases. The quartile method is then used to remove extreme values ​​from the weight allocation results, with the quartile range... Where Q1 is the lower quartile and Q3 is the upper quartile, exceeding... Extreme values ​​within a range are identified and removed. Simultaneously, the matching degree between the weight allocation results and the equipment and doctor resources of the health check center is verified. Specific verification indicators include whether the availability rate of equipment corresponding to high-weight health check items is ≥80% and whether the doctor's professional matching degree is ≥0.7. If these conditions are not met, the weight of the corresponding health check item is reduced, and the weight of alternative health check items is increased simultaneously. Finally, personalized adaptation parameters containing health check item weights and data collection priorities are formed. The data collection priority is positively correlated with the health check item weight; the higher the weight of the health check item, the higher the monitoring data collection priority.

[0027] Step S3 initiates a multi-dimensional data acquisition and preprocessing process based on the personalized adaptation parameters. First, S3-1, targeted data collection, is executed. The smart terminal initiates a data request to the historical database of physical examination users via HTTP / HTTPS protocol. The request message carries the user's identity identifier and data collection priority information. The historical database matches and queries the first data from the previous year's physical examination based on the identity identifier, sorts it according to collection priority, and returns it to the smart terminal. The first data includes structured data such as the test results, abnormality markers, and doctor's suggestions for each physical examination item in the previous year. Simultaneously, the smart terminal calls the intelligent physical examination order recommendation module via RPC protocol to obtain the second data from the physical examination recommendation order information table. The second data includes the basic execution order of the physical examination items, the dependencies between the various physical examination items, and the distribution information of the physical examination items in each region. For the collection of the third data, the smart terminal calls the open API interface of the mobile health APP and sports APP to collect data in a targeted manner according to the data collection priority. During the collection process, user authorization is obtained first. After authorization is granted, heart rate monitoring data, sleep time data, and various exercise data from the previous year to the date of the physical examination are collected. Among them, exercise data includes indicators such as exercise type, exercise duration, exercise intensity, and exercise frequency. The collection frequency is dynamically adjusted according to the data type. Heart rate data is collected every 5 minutes, sleep data is collected and summarized daily, and exercise data is collected once and synchronized in real time.

[0028] After data collection, S3-2 data standardization processing is performed. The Z-score standardization algorithm is used to clean the first, second, and third data sets. The Z-score standardization formula is as follows: Where X represents the original data, μ represents the mean of the data samples, and σ represents the standard deviation of the data samples, this algorithm transforms data of different dimensions into dimensionless standardized data, facilitating subsequent model analysis. During data cleaning, filtering rules are set according to the reasonable value range of each data type. The reasonable range for heart rate data is 30-200 beats / minute; data exceeding this range is marked as abnormal and filtered. The reasonable range for sleep time data is 3-12 hours; data below 3 hours or above 12 hours is considered invalid and removed. For exercise data, data with exercise intensity lower than 1.2 times the basal metabolic rate is considered low-intensity invalid exercise data and is filtered. For exercise app behavior logs, key exercise events are extracted using a behavior sequence analysis algorithm and transformed into standardized activity volume indicators, calculated as follows: ,in Let be the duration of the j-th motion. Let be the intensity coefficient of the j-th exercise (1.0 for resting state, 1.3-1.5 for low-intensity exercise, 1.6-1.9 for moderate-intensity exercise, and 2.0 and above for high-intensity exercise). For unstructured data from health apps, the TF-IDF algorithm is used to transform user operation preferences, search records, etc., into structured user health attention tags. The TF-IDF calculation formula is: ,in Let N be the frequency of the i-th keyword in the j-th record, and N be the total number of records. The number of records containing the i-th keyword is used to form a standardized multi-source dataset. The dataset is stored in CSV format, with fields including data type, standardized value, collection time, and data reliability.

[0029] Step S4 performs a large-scale dynamic comprehensive analysis based on the standardized multi-source dataset. First, it executes S4-1 multimodal data fusion modeling, employing a self-attention mechanism to fuse structured and unstructured data features from the standardized multi-source dataset to construct a multimodal fusion model. Structured data includes standardized quantitative indicators from the first, second, and third datasets; unstructured data includes user health concern tags, doctor's historical advice text, etc. The self-attention mechanism generates an attention weight matrix by calculating the correlation between each data feature and all other features. The calculation formula is as follows: Where Q is the query matrix, K is the key matrix, and V is the value matrix. The dimension of the key vector is d. Structured data is transformed into feature vectors of dimension d, and unstructured data is transformed into feature vectors of the same dimension through word embedding technology. The input is fed into the self-attention mechanism module to calculate attention weights. The feature vectors are weighted and summed according to the weights to obtain the fused multimodal feature vector. During the fusion process, the initial weight of the structured data feature vector is set to 0.6, and the initial weight of the unstructured data feature vector is set to 0.4. The final weights are dynamically adjusted through the attention mechanism.

[0030] Subsequently, real-time iterative optimization of the S4-2 prediction model is performed, and the model parameters are updated based on the improved incremental learning algorithm. The parameter update formula is as follows: ,in For the updated model parameters, These are the current model parameters. These are the parameters from the previous model round. The base learning rate ranges from 0.002 to 0.012 and is dynamically adjusted based on the data update frequency. When the data update frequency is higher than once per minute, When the value is between 0.002 and 0.005, and the data update frequency is less than once per minute, Take a value between 0.008 and 0.012; This is a data credibility weighting factor, ranging from 0.8 to 1.0. It is dynamically assigned based on the credibility of the data source; data collected by medical examination equipment has the highest credibility. The values ​​are assigned to 0.95-1.0, based on data collected by the mobile health app. The data was assigned a value of 0.85-0.9 and was collected by the fitness app. Assign a value of 0.8-0.85; Based on current parameters and real-time collected data The gradient of the loss function is calculated, using the mean squared error loss function, and the calculation formula is as follows: Where m is the number of data samples, This represents the actual label value (marking of abnormal items in historical physical examinations). These are the model's predicted values; This is a momentum factor, ranging from 0.05 to 0.15, used to accelerate parameter convergence and suppress oscillations. It is increased when the rate of decrease of the model loss function falls below a threshold. The value is set to 0.12-0.15. When the oscillation amplitude of the loss function exceeds the threshold, the value is reduced. Values ​​range from 0.05 to 0.08.

[0031] After the model parameters are updated, the S4-3 hierarchical results are accurately output, and the analysis results are classified according to the urgency of health risks and user awareness. The urgency of health risks is determined based on the risk probability predicted by the model. Health checkup items with a risk probability ≥ 0.7 are classified as high-risk items, those with a risk probability ≤ 0.3 and < 0.7 are classified as medium-risk items, and those with a risk probability < 0.3 are classified as low-risk items. User awareness is determined by analyzing users' historical health checkup feedback data and health knowledge query records. An awareness scoring algorithm is used to generate an awareness score of 0-10. A score ≥ 7 is classified as high awareness, 3 ≤ score < 7 is classified as medium awareness, and a score < 3 is classified as low awareness. High-risk items receive priority push notifications, with a doctor intervention reminder highlighted in red, and are simultaneously sent to the doctor's terminal at the health check center. Medium-risk items include a risk explanation text, tailored to the user's level of understanding: professional medical terminology for high-awareness users, and simplified explanations for low-awareness users. Low-risk items include personalized optimization suggestions based on the multi-dimensional user profile, generated using monitoring data from the third-party database. For example, for users with insufficient sleep, it's suggested to adjust exercise time to 3-5 PM to avoid evening exercise affecting sleep. The final optimized first result includes the health check item ID, risk level, predicted result, push priority, and additional information, including doctor reminders, risk explanations, and optimization suggestions.

[0032] Step S5, during the physical examination process, implements real-time adjustments and health warnings based on the optimized first result. First, it executes S5-1, dynamically optimizing the physical examination sequence. The smart terminal collects the user's physical status data in real time via Bluetooth Low Energy technology, including heart rate, steps, and fatigue score. The fatigue score is calculated by combining the heart rate coefficient of variation and the step count trend, using the following formula: HRV is the heart rate variability coefficient, and StepVar is the step count variability coefficient. Simultaneously, IoT sensors deployed in the health check center collect real-time pedestrian flow data in various areas. This data is obtained by combining the frequency of device usage within the area with video surveillance headcount statistics, and is updated every 2 minutes. Based on the weights of the health check items in the optimized first result, a greedy algorithm dynamically adjusts the order of health check execution. Each time, the health check item that is not currently being executed, has the highest weight, is suitable for the user's physical condition, and has low pedestrian flow in the area is selected as the next execution item. When the user's fatigue score is ≥6 points (out of 10), non-invasive and rapid health check items, such as height and weight measurement and blood pressure testing, are prioritized, while invasive or time-consuming health check items, such as venous blood sampling and abdominal CT scans, are postponed. When the pedestrian density in the area of ​​a popular health check item exceeds a threshold, a diversion mechanism is activated, adjusting some users to idle areas of similar devices or delaying their execution time to ensure a smooth health check process. In this embodiment, the threshold is 2 people per square meter.

[0033] Simultaneously, the S5-2 health warning system is implemented during the physical examination process, establishing a real-time communication link between the smart terminal and the examination equipment. This link utilizes the WebSocket protocol to ensure real-time data transmission and bidirectional interaction. After each test is completed, the examination equipment automatically synchronizes real-time vital sign data to the smart terminal via the communication link. The smart terminal compares this real-time vital sign data with preset thresholds. These preset thresholds are divided into general medical thresholds and personalized thresholds. The general medical threshold is set based on health standards, while the personalized threshold is dynamically adjusted based on the user's previous year's physical examination data and monitoring data from a third-party database. The calculation formula is as follows: ,in For general medical thresholds, The average of the user's historical data. This is a weighting coefficient, ranging from 0.6 to 0.8. When real-time vital signs data exceed the preset threshold by 10%-20%, a yellow alert is triggered, prompting the user to rest via a pop-up window on the smart terminal, while simultaneously recording the alert information. When real-time vital signs data exceed the preset threshold by 20% or more, a red alert is triggered. The smart terminal immediately synchronizes the alert information to the doctor's terminal at the health check center, displaying a pop-up window on the doctor's terminal along with the user's identity information, current examination items, and abnormal vital signs data, facilitating timely intervention by the doctor. The alert information is also simultaneously stored in the user's health record as reference data for subsequent analysis.

[0034] In addition, the S5-3 system enables real-time user feedback and rapid response. The smart terminal displays a concise questionnaire every 30 minutes, including a comfort rating (1-5 points) and an emergency request input box. Users can quickly check the rating or enter their emergency needs, such as urinary discomfort or dizziness. The smart terminal receives user feedback in real time and uses natural language processing algorithms to semantically analyze the emergency request text, extracting key demands. The analysis process uses a Naive Bayes classifier to categorize the request text into types such as process adjustment needs, medical assistance needs, and environmental adaptation needs. Based on the feedback, a process adjustment plan is generated within 10 seconds. For example, if a user reports urinary discomfort, the priority of the urinalysis test is elevated to the highest level and synchronized to the health check center's health check process management module. The process management module updates the user's health check execution plan and notifies the user of the adjusted examination order through electronic display and voice announcement. For medical assistance needs, such as dizziness or palpitations, a red alert mechanism is triggered to notify the doctor to check promptly. After the adjustment plan is implemented, the smart terminal displays the adjustment result in a pop-up window to ensure the user is aware of it.

[0035] Step S6 involves comparison correction and feedback iteration after all physical examination items are completed. First, S6-1, comparison correction of physical examination results, is performed. The physical examination center system collects the fourth data of this physical examination result. This fourth data includes the final test results of each physical examination item, abnormal markers, doctor's diagnostic opinions, etc., and is transmitted to the smart terminal through a data synchronization interface. The smart terminal associates and matches the fourth data with the optimized first result according to the physical examination item ID, compares the predicted results with the actual test results dimension by dimension, and calculates the deviation value. The deviation value includes absolute deviation and relative deviation. The formula for calculating the absolute deviation is... The formula for calculating the relative deviation is: Where Actual represents the actual detection value in the fourth dataset, and Predict represents the predicted value in the optimized first result. For health check items with large deviations, the causes of deviations are analyzed. Relative deviations ≥30% include incomplete data collection, model parameter deviations, and sudden changes in user health status. A moving average method is used to correct the model output error, and the correction formula is as follows: ,in This is a correction factor, ranging from 0.3 to 0.5, adjusted according to the cause of the deviation. Deviations caused by incomplete data collection are adjusted accordingly. The deviation caused by the model parameter deviation is taken Deviation caused by sudden changes in user health status .

[0036] Subsequently, S6-2 user profile and model parameter iterative update is performed. Anomalies in the fourth set of data are extracted as health risk tags and incorporated into the user health risk tag system. The tag system adopts a hierarchical structure: first-level tags include chronic disease risk, tumor risk, metabolic risk, etc.; second-level tags are specific disease risks, such as diabetes risk, lung cancer risk, etc.; and third-level tags are risk levels, including high, medium, and low. The corrected deviation data is then added to the user data sensitivity parameter. The data sensitivity parameter is used to adjust the frequency and weight of subsequent data collection. For example, if a certain type of data causes a large deviation, the collection frequency and reliability weight of that type of data are increased. Based on the corrected deviation... The parameters and weight factors of the prediction model are updated using batch gradient descent. The standardized multi-source dataset from this physical examination and the fourth dataset are combined to form a training sample set, which is then input into the model for secondary training to adjust the model's feature weights and bias terms. Simultaneously, the fourth dataset from this physical examination, the optimized first result, and the comparison and correction results are updated to the historical database of the physical examination users. The update strategy adopts incremental updates, adding only newly generated data records without overwriting historical data to ensure the integrity and traceability of the database. The updated multi-dimensional user profile and model parameters will be directly applied to the generation of personalized adaptation parameters for the user's next physical examination, achieving closed-loop iterative optimization throughout the entire process.

[0037] Example 3 The technical difference between this embodiment and Embodiment 1 is that this embodiment takes type 2 diabetes patients as an example. In order to target the physical examination scenario of chronic disease patients, while maintaining the core technical solution consistent with Embodiment 1, it refines the technology in a targeted manner based on the health management needs of chronic disease patients, so as to realize the personalized and precise optimization of physical examination for chronic disease patients.

[0038] In step S1, the user completes the physical examination registration through the online registration platform of the physical examination center. During the registration process, the system automatically fills in some static basic information by associating the user's medical insurance information and chronic disease management file with the ID card number, including the time of diagnosis of diabetes, current medication status, and history of past complications, reducing the user's manual input. The basic information supplementation interface displayed on the smart terminal adds special fields for diabetic users, including the frequency of blood glucose monitoring in the past 3 months, insulin usage, and dietary control adherence. The health demand collection module presets core demand options related to diabetes, including blood glucose control assessment, complication screening, medication adjustment suggestions, and diet and exercise guidance. Users can select multiple options and add custom demands, such as assessing the risk of fundus lesions and adjusting the dosage of oral hypoglycemic drugs. The static data and health demands together constitute a special basic information package for diabetic users. The data package additionally adds a chronic disease identification field and a chronic disease management level field, which are divided into 1-3 levels according to the years since diagnosis and the status of complications. Level 1 is newly diagnosed without complications, and Level 3 is diagnosed more than 5 years ago with complications.

[0039] In step S2, during the precise construction of the multi-dimensional user profile in S2-1, the modeling of metabolic indicators and complication risk dimensions is enhanced for diabetic users. In addition to the standard health risk score, a diabetes control score is added to the health status dimension, calculated using the following formula: ,in To determine the rate of achieving target blood glucose control, the percentage of days in the past three months with fasting blood glucose ≤7.0 mmol / L is used. For medication adherence scoring, For dietary control adherence scores, The system scores the risk of complications; the risk propensity dimension highlights the risk of diabetes-related complications, including the risk of retinopathy, nephropathy, neuropathy, and cardiovascular disease. Risk labels are generated by associating the user's diagnosis years, blood glucose control, blood pressure, and blood lipid data, such as high risk of retinopathy and moderate risk of nephropathy; the demand-oriented dimension directly maps the user's selected diabetes-related health needs, forming core demand labels for blood glucose assessment, complication screening, and medication adjustment.

[0040] In the S2-2 physical examination item priority dynamic weight allocation process, the weight factor values ​​are adjusted based on the multi-dimensional user profile of diabetic users. The matching degree coefficient weight is set to 0.55. The correlation coefficient weight is set to 0.3. The interaction weight was set to 0.15 to enhance the impact of the matching degree between user profile and physical examination items; In the process of calculating the matching coefficient between physical examination items and user profiles, special label weights are set for core physical examination items for diabetes, such as fasting blood glucose, glycated hemoglobin, routine urine test, fundus examination, kidney function test, and carotid ultrasound. For example, the diabetes-related label weight for the glycated hemoglobin physical examination item is set to 0.9, which is higher than the label weight for routine physical examination items. When calculating the correlation coefficient with abnormal data from the previous year, the focus is on exploring the relationship between data such as abnormal blood glucose, elevated urinary microalbumin, and abnormal fundus vessels from the previous year and the current physical examination items. For example, when urinary microalbumin was abnormal in the previous year, the correlation between renal function tests and renal ultrasound was significant. The value increased to 0.9; In the resource matching coefficient calculation process, priority is given to ensuring the compatibility of diabetes-specific physical examination equipment. For example, the availability weight of fundus photography equipment and glycated hemoglobin analyzers is increased to 0.5 to ensure the resource supply for high-weight physical examination items. This is achieved through a weight allocation formula. After calculation, the priority weight of core health check items for diabetes is generally higher than that of routine health check items, with glycated hemoglobin, fasting blood glucose, and fundus examination ranking in the top three.

[0041] During the real-time rationality verification of the S2-3 adaptation parameters, the screening criteria for similar user cases were expanded to include a requirement that the diabetes management level be consistent and the difference in blood glucose control scores be ≤2 points, ensuring the relevance of the cases. After removing extreme values, an additional verification of the weight ratio of chronic disease examination items was added, requiring that the sum of the weights of core diabetes examination items be ≥60% of the total weight. If this requirement is not met, the weight factors are adjusted and recalculated. The resource matching degree verification involves a special inspection of diabetes-specific equipment to ensure that the availability rate of glycated hemoglobin analyzers, fundus photography equipment, and urine microalbumin analyzers is ≥90%. If a specific piece of equipment malfunctions, the corresponding examination item is automatically replaced with an alternative solution. For example, if the fundus photography equipment malfunctions, it is replaced with fundus endoscopy, while reducing the weight decay of the alternative solutions to ensure that chronic disease testing needs are met. In the final generated personalized adaptation parameters, the data collection priority is set to the highest priority for diabetes-related monitoring data, such as blood glucose monitoring data, insulin usage records, and dietary intake data, with the collection frequency increased to once every 3 minutes.

[0042] In step S3, during the targeted data collection process (S3-1), in addition to the regular first, second, and third data, additional data specific to the diabetes management app is collected, including fasting blood glucose data for the past 3 months, 2-hour postprandial blood glucose data, insulin injection records, dietary diaries, and blood glucose fluctuation curves. By connecting to the hospital's chronic disease management system, data such as the user's glycated hemoglobin test results, complication screening records, and doctor's treatment recommendations for the past year are obtained and added to the first data. During the third data collection, exercise data focuses on the duration and frequency of moderate-intensity aerobic exercise, such as brisk walking, jogging, and swimming. Health app data focuses on the response rate of blood glucose monitoring reminders and health knowledge learning records, such as knowledge related to diabetes diet and exercise. During data collection, an encrypted transmission protocol, such as TLS 1.3, is used for blood glucose data to ensure the security of sensitive medical data. After collection, the diabetes-specific data is separately labeled to facilitate subsequent targeted analysis by the model.

[0043] During the S3-2 data standardization process, specific cleaning rules were established for diabetes-specific data. The reasonable range for fasting blood glucose data is 3.9-11.1 mmol / L, and the reasonable range for 2-hour postprandial blood glucose data is 3.9-13.9 mmol / L. Data exceeding these ranges needs to be assessed for validity in conjunction with medication use and dietary records. For example, if a user's blood glucose level is below 3.9 mmol / L after insulin injection but without hypoglycemic symptoms, this data is considered valid and retained. The reasonable range for glycated hemoglobin data is 4.0%-10.0%, and data exceeding this range is filtered. For blood glucose fluctuation curve data, a sliding window algorithm is used to extract features such as fluctuation amplitude, peak value, and trough value, converting them into standardized fluctuation indicators. The calculation formula is as follows: Where Peak is the peak blood glucose level and Trough is the trough blood glucose level. The average blood glucose level was used. For food diary data, a nutrient composition analysis algorithm was used to extract indicators such as carbohydrate intake and dietary fiber intake, which were then converted into standardized nutrient intake coefficients to ensure that the standardized processing of diabetes-specific data met the needs of subsequent model analysis.

[0044] In step S4, during the multimodal data fusion modeling process in S4-1, a metabolic feature submodule is added for diabetic users. Blood glucose data, glycated hemoglobin data, insulin usage data, dietary data, etc., are integrated into a metabolic feature vector, which is input into the self-attention mechanism module along with the feature vectors of other data modalities. When calculating the self-attention mechanism, the initial attention weight of the metabolic feature vector is set to 0.7, which is higher than that of the conventional feature vector, to ensure that chronic disease-related data plays a dominant role in the fusion process. In the fused multimodal feature vector, the proportion of diabetes-related feature dimensions increases to 40%, strengthening the model's focus on chronic disease indicators.

[0045] During the real-time iterative optimization of the S4-2 prediction model, the data credibility weight factor... Adjustments were made to the diabetes-specific data provided by the hospital's chronic disease management system. The value was assigned to 0.98, based on data from a diabetes management app. The value was assigned to 0.92, based on data from the blood glucose monitoring device. The value was assigned to 0.99 to ensure that highly reliable chronic disease data contributes more to model updates; when calculating the loss function, the prediction error of the core health check items for diabetes was given higher weight, and the calculation formula was adjusted to... ,in This involves a set of core health checkup items for diabetes. By increasing the error weights of these core items, the model is guided to focus on optimizing predictions related to chronic diseases. (Base learning rate) When updating diabetes user data, the value is set to 0.005-0.008 to balance iteration speed and prediction accuracy.

[0046] In the S4-3 hierarchical results output process, the criteria for high-risk items have been adjusted for diabetic users. A risk probability ≥0.6 for blood glucose-related indicators is considered high-risk, as are complications-related examination items such as fundus examinations and kidney function tests with a risk probability ≥0.5. Push notifications now include specific content for diabetic users. High-risk items' doctor intervention reminders are accompanied by a priority list of diabetes specialists, ensuring priority allocation to doctors specializing in the diagnosis and treatment of diabetic complications. Medium-risk items' risk interpretation text incorporates the pathological mechanisms of diabetes; for example, elevated glycated hemoglobin suggests poor blood glucose control over the past 3 months, and long-term hyperglycemia may increase the risk of fundus lesions. Personalized optimization suggestions for low-risk items are generated based on diabetes management needs; for example, exercise data shows insufficient moderate-intensity exercise, suggesting adding 3 brisk walks per week for 30 minutes each time, avoiding exercise on an empty stomach, and monitoring blood glucose 1 hour after exercise. The optimized first result includes an additional diabetes management suggestion field, summarizing optimization suggestions for each examination item to form a systematic chronic disease management plan.

[0047] In step S5, during the dynamic optimization of the physical examination sequence in S5-1, fasting examination items, such as fasting blood glucose, glycated hemoglobin, liver function, and kidney function, are prioritized based on the fasting needs and physical tolerance characteristics of diabetic users. These are concentrated and completed within one hour after the start of the physical examination to avoid hypoglycemia caused by prolonged fasting. When collecting user physical status data, real-time blood glucose monitoring is added. A Bluetooth blood glucose meter is connected to a smart terminal, and fingertip blood glucose data is collected every 15 minutes. When blood glucose is lower than 3.9 mmol / L, the process adjustment is immediately triggered, the current physical examination item is paused, a hypoglycemia warning is pushed and sugar supplementation is recommended. Once blood glucose recovers to above 4.4 mmol / L, non-invasive physical examination items are prioritized, and invasive physical examination items are postponed. When counting the flow of people, separate statistics are collected for diabetic-specific physical examination areas, such as the fundus examination room and chronic disease treatment area, to ensure that the flow density in these areas is less than 1.5 people per square meter, reducing user waiting time and the risk of cross-infection.

[0048] During the health alert process of the S5-2 physical examination, the preset thresholds are personalized for diabetic users. The general medical threshold for fasting blood glucose is 3.9-6.1 mmol / L, and the personalized threshold calculation formula is adjusted as follows: ,in The system uses the user's average fasting blood glucose level over the past three months. For example, if the user's average fasting blood glucose level over the past three months is 6.5 mmol / L, then the personalized fasting blood glucose threshold lower limit is adjusted to 3.9 mmol / L, and the upper limit is adjusted to 6.5 + (6.1 - 3.9) × 0.3 = 7.16 mmol / L. The warning mechanism adds a special warning for hypoglycemia. When the real-time blood glucose level is lower than 3.9 mmol / L, an orange warning is triggered. In addition to a pop-up reminder, the smart terminal simultaneously sends a request for help to the nursing station of the physical examination center, and the nurses bring sugary food to intervene in time. When the blood glucose level is lower than 3.0 mmol / L, a red warning is triggered. All physical examination items are immediately suspended, and the doctor is notified for emergency treatment. At the same time, the user's chronic disease management file is accessed to obtain the history of hypoglycemia treatment, which is provided to the doctor for reference.

[0049] During the S5-3 user real-time feedback and rapid response process, the pop-up questionnaire adds a special feedback field for diabetic users, including current blood sugar levels (no discomfort, mild hunger, significant hunger, dizziness, palpitations); tolerance to physical examination items (no discomfort, mild discomfort, significant discomfort); when users report significant hunger or dizziness and palpitations, the system automatically identifies it as suspected hypoglycemia symptoms and immediately initiates the hypoglycemia warning process without waiting for the user to input detailed requirements; when adjusting the plan, the safety and comfort of diabetic users are prioritized. For example, if a user reports arm discomfort after venous blood collection, subsequent physical examination items requiring blood collection can be postponed to the next day or replaced with a non-invasive testing plan, and the results are simultaneously sent to the doctor's terminal for the doctor to assess the feasibility of the testing plan.

[0050] In step S6, during the physical examination result comparison and correction process in S6-1, the deviation analysis for the core physical examination items for diabetes focuses on blood glucose-related indicators, analyzing whether the deviation is caused by factors such as dietary changes, medication adjustments, and exercise interventions; during deviation correction, for blood glucose prediction deviations, a diabetes-specific correction model is used, and the correction formula is adjusted as follows: ,in The system uses blood glucose trend values ​​to improve correction accuracy. By comparing the correction results, an additional diabetes index deviation report is generated, recording the predicted value, actual value, deviation value, and cause of deviation for each blood glucose-related index, providing a basis for subsequent model optimization.

[0051] During the iterative update of the S6-2 user profile and model parameters, diabetes-related abnormalities in the fourth data set, such as elevated glycated hemoglobin, positive urinary microalbumin, and retinal vascular stenosis, were updated to the user health risk labeling system. New early warning labels for complications were added, such as early kidney damage risk and retinal vascular sclerosis risk. User data sensitivity parameters were adjusted for blood glucose and insulin usage data, increasing the collection frequency and model weight of these data types. When updating the prediction model parameters, a diabetes-specific training set was used for fine-tuning. This training set included historical physical examination data, monitoring data, and correction results for similar diabetes users. A transfer learning algorithm was used to adapt the general model to a diabetes-specific model. When updating the historical database of physical examination users, a separate diabetes user sub-database was established, categorized and stored according to management level, complication status, and blood glucose control level, facilitating subsequent screening of similar user cases and model training. The updated user profile and model parameters will be directly applied to the user's next diabetes-specific physical examination, achieving closed-loop optimization of the entire chronic disease management cycle.

[0052] Example 4 The technical difference between this embodiment and Embodiment 1 is that this embodiment takes internet industry practitioners aged 25-45 as an example to target the physical examination scenario of working professionals. Combining the characteristics of working professionals such as fast work pace, high pressure, irregular work and rest, and tight physical examination time, this embodiment makes targeted optimizations based on the core technical solution to achieve efficient, accurate and convenient physical examination services for working professionals.

[0053] In step S1, the physical examination registration process supports two modes: bulk registration for corporate group examinations and quick registration for individuals. For corporate group examinations, the company's HR uploads employee identity information and physical examination package information, and employees complete the supplement of personal information via SMS link, reducing registration time. Quick registration for individuals supports Alipay and WeChat quick login, linking electronic ID cards and health codes to achieve one-click registration. The basic information supplementation interface adds occupation-related fields for working professionals, including years of service, average daily working hours, overtime frequency, work mode, and commuting method. The health demand collection module presets high-frequency demand options for working professionals, including sub-health status assessment, cervical and lumbar spine examination, cardiovascular risk screening, sleep quality assessment, and stress management suggestions. Users can quickly select and supplement custom demands, such as screening for fatty liver and assessing the cause of vision loss. The basic information package also adds a physical examination time preference field, allowing users to select the physical examination time period and the expected examination duration, providing a basis for subsequent process optimization.

[0054] In step S2, during the precise construction of the multi-dimensional user profile in S2-1, the modeling of workplace-related health risk dimensions is strengthened, and a workplace sub-health score is added to the health status dimension. The calculation formula is as follows: ,in A sleep quality score is given on a scale of 0-10, calculated based on sleep duration and sleep continuity. A stress score is given on a scale of 0-10, based on overtime frequency and working hours. Posture is scored on a scale of 0-10, based on work patterns and historical data on the cervical and lumbar spine. Rate the sport on a scale of 0-10. Dietary scores are assigned from 0 to 10. The risk propensity dimension highlights the risk of common workplace diseases, including cervical spondylosis, lumbar disc herniation, fatty liver, hypertension, and neurasthenia. Risk labels are generated by associating work patterns, overtime frequency, exercise data, and dietary data, such as high risk of cervical spondylosis and moderate risk of fatty liver. The demand-oriented dimension combines users' preferences for physical examination time to form core demand labels for efficient physical examinations, key screenings, and convenient suggestions.

[0055] In the S2-2 physical examination item priority dynamic weight allocation process, the weight factors are adjusted to meet the efficient physical examination needs of working professionals. Set it to 0.45. Set it to 0.35. Set it to 0.2 to increase the weight of the correlation coefficient of historical abnormal data; When calculating, the weight of labels for common workplace diseases-related medical examination items such as cervical spine DR, lumbar spine DR, abdominal ultrasound, blood pressure, blood lipids, liver function, thyroid function, and sleep monitoring is increased. For example, the workplace-related label weight for cervical spine DR is set to 0.85. When calculating, the focus is on exploring the correlation between data from the previous year, such as cervical spine abnormalities, lumbar spine abnormalities, elevated blood lipids, elevated blood pressure, and fatty liver, and the current physical examination items. When calculating, priority should be given to the testing time of each physical examination item. Items with a testing time of ≤10 minutes should be considered first. The value is increased by 0.1 for physical examination items with a testing time of ≥30 minutes. The value is reduced by an additional 0.1 to guide the weighting towards rapid testing items. After calculation using the weighting allocation formula, the priority weights of physical examination items related to common workplace diseases and rapid testing items are significantly increased, forming a ranking of physical examination items that can quickly complete core screening.

[0056] During the real-time rationality verification of the S2-3 adaptation parameters, the selection criteria for similar user cases were expanded to include consistent work patterns, similar overtime frequencies, and an age range of ±3 years, ensuring that the cases closely match the characteristics of working professionals. After removing extreme values, a verification of the total examination time was added, requiring that the total examination time of the weighted examination items be ≤ the user's expected examination time + 30 minutes. If this is exceeded, low-weight examination items are automatically optimized and replaced with faster alternatives or postponed to the next examination. Resource matching verification specifically examines rapid testing devices such as portable blood pressure monitors and rapid blood lipid analyzers, as well as high-frequency examination items such as cervical spine DR machines and abdominal ultrasound machines, ensuring that the turnover efficiency of these devices is ≥90% to reduce user waiting time. In the final generated personalized adaptation parameters, data collection priority is set high for high-frequency monitoring data of working professionals, such as exercise data, sleep data, and stress-related data, while optimizing the timing of data collection to avoid frequently disturbing users during work hours.

[0057] In step S3, during the targeted data collection process in S3-1, in addition to regular data, additional workplace-related data is collected, including office software usage time, commuting time, overtime records, and food delivery order records. During the third data collection phase, exercise data focuses on collecting the duration of fragmented exercise and concentrated weekend exercise; sleep data focuses on collecting the difference in sleep between weekdays and weekends; and health app data focuses on collecting learning records of workplace health knowledge. Data collection supports offline caching and batch upload modes. Data collected during work hours is temporarily stored locally and automatically uploaded in batches during non-work hours to reduce work interference. The collection frequency is adjusted for the working population, with the data collection frequency reduced to once every hour during work hours and returning to the normal collection frequency during non-work hours.

[0058] During the S3-2 data standardization process, specific rules were established for workplace-specific data. Office software usage time was standardized based on daily average duration, calculated using the following formula: ,in This refers to the average daily usage time of users' office software. The minimum value for users of the same type. The data represents the maximum value for similar users; food delivery records are analyzed using a nutritional composition algorithm to extract the ordering frequency of high-oil, high-salt, and high-sugar foods, converting it into a dietary health coefficient; fragmented exercise data is weighted by cumulative duration and intensity to calculate standardized activity levels, ensuring the value of fragmented exercise for working professionals is reflected; during data cleaning, interpolation is used to supplement missing data for working professionals during their work hours. For example, if exercise data for a certain workday is missing, it is supplemented by the average of the user's recent exercise data for similar workdays and weekends, ensuring data completeness.

[0059] In step S4, during the multimodal data fusion modeling process in S4-1, a workplace feature submodule is added to integrate office hours data, commuting data, overtime records, and food delivery data into a workplace feature vector, which is then fused with other data modalities. When calculating the self-attention mechanism, the initial attention weights of both the workplace feature vector and the health feature vector are set to 0.5 to balance the influence of workplace factors and health factors. In the fused multimodal feature vector, workplace-related features account for 35%, health-related features account for 35%, historical physical examination data features account for 20%, and resource adaptation features account for 10%, ensuring that the model fully considers the special characteristics of the working population.

[0060] During the real-time iterative optimization of the S4-2 prediction model, the data credibility weight factor... In response to adjustments to workplace-specific data, the enterprise OA system provides overtime records and commuting data. The value was assigned to 0.9, based on data from the food delivery app. Assign a value of 0.8 to the office software usage time data. The base learning rate is assigned a value of 0.85. When calculating the loss function, the prediction error of common workplace health check-up items is weighted 1.3 times to guide the model to focus on optimizing the prediction accuracy of these items. Set the time to 0.008-0.012 during non-working hours on weekdays to complete the fasting physical examination item iteration during users' rest time, thus avoiding the occupation of working resources.

[0061] In the precise output of S4-3 hierarchical results, the criteria for high-risk items have been adjusted to address common workplace illnesses. Cervical spondylosis, lumbar disc herniation, fatty liver, and hypertension are classified as high-risk if the risk probability is ≥0.65. Push notifications are optimized for working professionals, using a concise format with core conclusions and brief suggestions. For high-risk items, doctor intervention reminders support appointments with specialist clinics, such as orthopedics or gastroenterology, with priority given to appointments within one week of the physical examination. For medium-risk items, the risk interpretation text is tailored to workplace scenarios, such as: prolonged sitting can straighten the lumbar spine's physiological curvature, suggesting getting up and moving around for 5 minutes every hour. For low-risk items, personalized optimization suggestions are generated based on the time constraints of working professionals, such as: fragmented exercise suggestions, such as getting off one stop earlier during commutes and walking, and performing 5 minutes of neck stretches every hour in the office. The optimized first result supports PDF export, making it easy for users to save and submit to company HR for record-keeping.

[0062] In step S5, during the dynamic optimization of the physical examination order in S5-1, a centralized and efficient optimization strategy is adopted based on the time arrangements of working professionals. Physical examination items of the same type and location are grouped together to reduce the distance users travel within the physical examination center. For example, all imaging examinations are performed consecutively on the same floor, and all blood-related examinations are completed at a single window. When collecting user physical condition data, stress level monitoring is added. Based on the heart rate variability coefficient, when the stress level score is ≥7, a 5-minute relaxation session is inserted, such as providing meditation audio or a massage chair for rest, before continuing with subsequent examinations. The flow of people data supports real-time appointment functionality. Users can view the current number of people waiting for each examination item and the estimated waiting time through their smart terminals, and choose whether to adjust the examination order. For example, if a certain examination item has a large number of people waiting, users can choose to perform other examinations with fewer people waiting first.

[0063] During the health warning process of the S5-2 physical examination, preset thresholds are adjusted for working professionals. The general medical threshold for blood pressure is 90 / 60-140 / 90 mmHg. Personalized thresholds are combined with workplace stress factors, and the calculation formula is as follows: ,in The system provides users with their average blood pressure over the past three months. A new warning mechanism has been added specifically for cervical and lumbar spine issues. When the medical examination equipment detects abnormal cervical curvature or a high risk of lumbar disc herniation, a blue warning is triggered, and targeted protection suggestions are pushed, such as avoiding prolonged periods of looking down at the computer and adjusting the monitor height and chair angle. For common workplace eye strain issues, real-time vision monitoring has been added. When uncorrected visual acuity decreases by ≥0.2 compared to the previous year, a yellow warning is triggered, recommending a detailed eye exam.

[0064] During the S5-3 user real-time feedback and rapid response process, the pop-up questionnaire adopts a one-click feedback design. Users can quickly provide feedback by clicking the icon, such as long waiting time, equipment incompatibility, or need to speed up the process. When a user reports that the waiting time is too long, the system automatically checks the waiting queue of the current physical examination item, assigns a priority channel to the user, or adjusts the order of subsequent physical examination items, replacing physical examination items with those with readily available equipment. When a user reports that the process needs to be expedited, the system automatically optimizes the process of subsequent physical examination items, merging repetitive steps, so that multiple examinations of the same part are completed in one go, ensuring that the physical examination is completed within the user's expected time.

[0065] In step S6, during the physical examination result comparison and correction process (S6-1), considering the unique characteristics of physical examination data for working professionals, the focus is on analyzing the impact of work-related factors on the examination results. For example, whether elevated blood lipids are related to long-term takeout consumption and lack of exercise, and whether cervical spine abnormalities are related to long hours of office work. During deviation correction, a workplace scenario correction model is used, and the correction coefficient is adjusted by combining data such as the user's work patterns and overtime frequency to improve the accuracy of the correction. The correction results are compared to generate a workplace health improvement report, which is presented in categories such as cervical spine protection, lumbar spine protection, dietary adjustments, exercise suggestions, and sleep optimization, making it convenient for users to make targeted improvements.

[0066] During the S6-2 user profile and model parameter iteration update process, workplace-related anomalies in the fourth data set were updated to the user health risk labeling system. New primary workplace-related health risk labels were added, including secondary labels such as office-related injury risks, diet-related metabolic risks, and work-rest-related sleep risks. User data sensitivity parameters were adjusted for workplace-specific data, increasing the model weights of office hours, overtime records, and food delivery data. When updating the prediction model parameters, workplace-specific training data was incorporated to optimize the model's predictive ability for common workplace diseases. When updating the historical database of health checkup users, data was categorized and stored according to occupational type and work mode, providing data support for subsequent health checkup optimization for similar workplace groups. The updated user profile and model parameters will be applied to the user's next health checkup, enabling continuous optimization of workplace health management. Simultaneously, it supports corporate HR in obtaining anonymized employee health checkup statistical reports, providing a basis for companies to formulate health welfare policies.

[0067] Example 5 The technical difference between this embodiment and Embodiment 1 is that this embodiment is designed for physical examinations of people aged 60 and above. It takes into account the characteristics of the elderly, such as declining physical function, difficulty in movement, weakened cognitive ability, and multiple underlying medical history. Based on the core technical solution, it optimizes for age-friendliness to improve the safety, comfort, and convenience of physical examinations for the elderly.

[0068] In step S1, the physical examination registration supports a children's proxy mode, where children can complete online registration and supplement basic information on behalf of their parents by binding their identity information, reducing the difficulty of operation for the elderly. The offline registration terminal is equipped with a large-font interface, voice navigation, and manual assistance functions, and staff can assist in completing the registration operation. The basic information supplementation interface adds a special basic medical history input module for the elderly, using a simplified question-and-answer format, such as: Do you have high blood pressure? Do you have diabetes? Do you have heart disease? It supports voice answers and head nodding and shaking recognition. The health demand collection module presets high-frequency demand options for the elderly, including basic disease control assessment, cardiovascular and cerebrovascular risk screening, joint function assessment, and cognitive function screening. Users or their children can select and add custom demands, such as assessing the cause of hearing loss and checking for osteoporosis. The basic information package additionally adds a mobility level field, which is divided into 1-3 levels according to independent walking ability and whether assistive devices are needed, with level 1 being fully independent walking and level 3 requiring wheelchair assistance, and a cognitive ability level field, which is divided into 1-3 levels according to communication ability and memory, with level 1 being normal cognition and level 3 being cognitive decline.

[0069] In step S2, during the accurate construction of the multi-dimensional user profile in S2-1, the modeling of the dimensions related to underlying diseases and aging is strengthened, and a comprehensive health score for the elderly is added to the health status dimension. The calculation formula is ,in Basic disease control score, Assess physical function. To score cognitive function, Assess action capabilities. Nutritional status is scored; the risk propensity dimension highlights the risk of common diseases among the elderly, including the risk of hypertension complications, coronary heart disease, stroke, osteoporosis, and cognitive impairment. Risk labels are generated by linking basic medical history, age, and physical function data, such as high stroke risk and moderate osteoporosis risk; the needs-oriented dimension combines the behavioral and cognitive characteristics of the elderly to form core needs labels that prioritize safety, focus on screening, and simplify procedures.

[0070] In the S2-2 physical examination item priority dynamic weight allocation process, the weight factors are adjusted according to the characteristics of the elderly. Set it to 0.5. Set it to 0.3. Set to 0.2 to enhance the matching degree between user profiles and physical examination items; When calculating, the label weights of physical examination items related to diseases with high incidence among the elderly, such as blood pressure, blood lipids, blood sugar, electrocardiogram, echocardiogram, carotid ultrasound, bone density test, and cognitive function assessment, were significantly increased. For example, the elderly-related label weight for bone density test was set to 0.9. During the calculation, the focus is on exploring the correlation between the previous year's data on hypertension, hyperlipidemia, hyperglycemia, and abnormal electrocardiograms and the current physical examination items; When calculating, priority should be given to the safety and comfort of the physical examination items, especially non-invasive ones. The value is increased by 0.15 for invasive physical examination items. The weighting is reduced by 0.1, and the weighting of the location convenience of the equipment for physical examination items for elderly people with limited mobility is increased to 0.4, ensuring that high-weight physical examination items are easily accessible to the elderly. After calculation using the weighting allocation formula, the priority of non-invasive, safe, and convenient physical examination items for common diseases among the elderly is significantly increased, while the weight of invasive physical examination items is appropriately reduced or scheduled for later execution.

[0071] During the real-time rationality verification of the S2-3 adaptation parameters, the screening criteria for similar user cases were expanded to include age ±5 years, consistent basic medical history, and the same level of mobility, ensuring that the cases closely match the physical condition of the elderly. After eliminating extreme values, a safety verification of the physical examination items was added, requiring that the risk level of high-weight physical examination items be ≤2, with level 1 being no risk, level 2 being low risk, level 3 being medium risk, and level 4 being high risk. If any physical examination item has a risk level of 3 or higher, risk disclosure and informed consent prompts must be added to the adaptation parameters. The resource matching degree verification specifically examines elderly-friendly equipment, such as large-print blood pressure monitors, height-adjustable examination beds, and low-radiation CT scanners, ensuring that the availability of these devices is ≥95%. At the same time, the accessibility facilities of the physical examination center are verified to be complete, such as ramps, elevators, and accessible restrooms. In the final generated personalized adaptation parameters, the data collection priority is set to the highest priority for monitoring data related to the elderly's basic diseases, such as blood pressure, blood sugar, and heart rate, and the collection frequency is adjusted to once every 10 minutes to ensure real-time monitoring of physical condition.

[0072] In step S3, during the targeted data collection process in S3-1, in addition to routine data, additional data related to elderly care is collected, including caregiver information, daily medication lists, and home environment information. During the third data collection phase, exercise data focuses on daily activity levels, such as indoor walking and outdoor strolls; sleep data focuses on nighttime awakenings and sleep continuity; and health app data focuses on medication reminder response rates and health monitoring adherence. Data collection supports a simplified operation mode, with the smart terminal interface featuring large fonts and high contrast. The collection process includes voice guidance, such as: "Please confirm your blood pressure data for today." For elderly individuals who do not use smart devices, data can be entered by their children or manually by staff through the physical examination center terminal to ensure the completeness of the data collection. The collected medication list data is correlated with the physical examination items to avoid conflicts between the items and medications. If certain physical examination items require discontinuation of medication, this must be communicated in advance.

[0073] During the S3-2 data standardization process, special rules were formulated for the characteristics of elderly data. For example, considering the greater fluctuations in blood pressure among the elderly, blood pressure data was standardized by taking the average of three measurements. The calculation formula is as follows: ,in , , The data includes blood pressure measurements at different time points; for exercise data, which is designed for older adults with lower activity levels, absolute activity levels are used for standardization rather than relative values ​​to ensure accurate quantification of even mild activity; during data cleaning, more lenient filtering rules are applied to abnormal data caused by cognitive or operational issues in older adults, and a comprehensive judgment is made based on historical data and underlying medical conditions to avoid mistakenly deleting valid data; for example, blood glucose data mistakenly recorded by older adults is considered valid and retained if it does not deviate significantly from historical data and is within the control range of underlying medical conditions.

[0074] In step S4, during the multimodal data fusion modeling process in S4-1, an elderly feature submodule is added to integrate basic disease data, medication data, care data, mobility data, etc., into an elderly feature vector, which is then fused with other data modalities. When calculating the self-attention mechanism, the initial attention weight of the elderly feature vector is set to 0.75 to ensure that the special needs of the elderly are given priority. In the fused multimodal feature vector, basic disease-related features account for 40%, safety-related features account for 25%, comfort-related features account for 20%, and resource adaptation features account for 15%, strengthening the model's emphasis on the health and safety of the elderly.

[0075] During the real-time iterative optimization of the S4-2 prediction model, the data credibility weight factor... Adjustments to data on the elderly and hospital visit data The value was assigned to 0.98, which is the data entered by the children. The value was assigned to 0.9, which is the data entered by the staff. A value of 0.95 is assigned to ensure the dominant role of high-confidence data in model updates; when calculating the loss function, the prediction errors of health check-up items for common diseases among the elderly and safety-related health check-up items are assigned a weight of 1.6 times, guiding the model to focus on optimizing the prediction accuracy of these types of health check-up items; the base learning rate... Set the value to 0.001-0.003 to reduce the iteration speed, ensure the stability of model parameters, and avoid large changes in prediction results due to data fluctuations.

[0076] In the S4-3 hierarchical results output process, considering the cognitive decline of the elderly, the output results adopt a combination of text and images, along with voice broadcast. The text uses large fonts and concise expressions, avoiding technical jargon, such as: "Your blood pressure is well controlled; please continue your current medication." For high-risk items, the doctor intervention reminder uses a flashing red animation and voice prompts, simultaneously displayed on the caregiver's terminal to ensure timely awareness. For medium-risk items, the risk interpretation text combines the elderly's underlying diseases and explains in layman's language, such as: "Your bone density is a bit low, making you prone to fractures; we recommend eating more calcium-rich foods, such as milk and tofu." For low-risk items, personalized optimization suggestions are generated based on the elderly's activity and dietary characteristics, such as: "Exercise suggestion: Take a 20-minute walk with family members after 10 am every day to avoid falls"; "Dietary suggestion: Cook food softer and easier to digest." The optimized first result can be printed as a large-print paper report and simultaneously sent to children's mobile phones for easy interpretation and follow-up care.

[0077] In step S5, during the dynamic optimization of the physical examination sequence in S5-1, a short-distance, minimal-movement optimization strategy is adopted to address the mobility limitations of the elderly. The examination items are arranged sequentially according to the examination center area to avoid cross-floor or long-distance travel. For example, examination items on the first floor are prioritized, followed by those on the second floor, reducing the number of times users need to go up and down stairs. When collecting user physical condition data, an additional physical exertion assessment is added, using a physical exertion scoring algorithm. When the score is ≥6, a rest period is arranged, such as sitting quietly in the rest area for 15 minutes and drinking warm water before continuing with subsequent examination items. When collecting traffic data, a green channel is set up for the elderly, allowing high-weight examination items to be prioritized without queuing. Furthermore, based on the elderly's daily routines, the examination time slot between 9-11 am is prioritized, as the elderly are in better physical condition during this period, avoiding examinations conducted too early or too late.

[0078] During the health warning process of the S5-2 physical examination, preset thresholds are adjusted individually for elderly individuals. The general medical threshold for blood pressure is 90 / 60-150 / 95 mmHg (standard for the elderly). The formula for calculating the personalized threshold is as follows: ,in The system displays the user's average blood pressure over the past three months to avoid triggering unnecessary warnings due to slight fluctuations in blood pressure. A safety-related warning mechanism has been added: a red warning is triggered when the medical examination equipment detects a heart rate of <55 beats / minute (too slow) or >110 beats / minute (too fast), immediately suspending the examination and notifying a doctor for further examination. When the elderly person moves within the medical examination center, the smart terminal monitors their movement speed via positioning; a yellow warning is triggered if the movement is too slow or the person stays in one place for too long, notifying staff to check if assistance is needed. Warning information is synchronized to the caregiver's terminal, ensuring that caregivers are aware of the elderly person's condition in real time.

[0079] During the S5-3 user real-time feedback and rapid response process, the pop-up questionnaire uses voice feedback and simple gestures. Seniors can express their discomfort, need water, or need to use the restroom by voice, or by clicking the corresponding icon on the screen to provide feedback. When a user reports discomfort, the system immediately triggers rapid collection of physical status data and notifies nearby staff to assist. Based on the feedback, the system adjusts the physical examination process, such as pausing the examination, providing medical assistance, or arranging rest. When a user reports needing to use the restroom, the smart terminal displays the location of the nearest accessible restroom and notifies staff to assist in guiding the senior to the restroom, ensuring the senior's safe arrival.

[0080] In step S6, during the S6-1 physical examination result comparison and correction process, considering the characteristics of elderly people having multiple underlying diseases and declining physical functions, the focus is on analyzing the comprehensive impact of underlying diseases and aging on the physical examination results. For example, whether elevated blood sugar is related to aging and underlying diabetes, and whether blood pressure fluctuations are related to cardiovascular function decline; during deviation correction, an elderly-specific correction model is used, and the correction coefficient is adjusted in combination with the user's underlying medical history, age, and medication status. The calculation formula is as follows: ,in The underlying disease impact factor is set to a value of -0.1 to 0.1 based on the type and control status of the underlying disease. The correction results are compared to generate a health care guide for the elderly, which is presented in categories such as medication guidance, dietary recommendations, exercise recommendations, home safety, and regular check-ups. The content is specific, operable, and easy for caregivers to implement.

[0081] During the iterative update of the S6-2 user profile and model parameters, age-related anomalies in the fourth data set were updated to the user health risk labeling system, such as decreased bone density, cognitive decline, and joint degeneration. New primary labels related to age-related decline were added, including secondary labels for physiological function decline risk, cognitive function decline risk, and mobility safety risk. User data sensitivity parameters were adjusted for underlying disease data and safety-related data of the elderly, increasing the collection frequency and model weight of these data. When updating the prediction model parameters, specific training data for the elderly population was added to optimize the model's predictive ability for common diseases and safety risks among the elderly. When updating the historical database of physical examination users, data was categorized and stored by age group and type of underlying medical history, providing data support for subsequent optimization of physical examinations for similar elderly populations. The updated user profile and model parameters will be applied to the user's next physical examination, achieving a closed-loop optimization of safe and comfortable health management for the elderly, while providing personalized care suggestions for caregivers and improving the quality of life for the elderly.

[0082] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for health information push and physical examination optimization, characterized in that, Includes the following steps: S1: The user completes the physical examination registration and basic information entry, performs the physical examination registration operation, and supplements static basic information and health requirements; S2: Generate personalized physical examination adaptation parameters, construct a user profile based on the basic information and health needs, assign priority weights to physical examination items and verify their rationality; S3: Smart terminal multi-dimensional data acquisition and preprocessing, acquire multi-source data according to the adaptation parameters and perform standardized processing to obtain standardized data; S4: The large model performs dynamic comprehensive analysis and outputs optimization results. It integrates and models the standardized data obtained in S3 and iteratively optimizes the prediction model, outputting analysis results and health information related to the current physical examination item in a hierarchical manner. S5: Real-time adjustments and health alerts during physical examinations, specifically including: S51: Real-time collection of physical status data of users undergoing physical examinations and traffic data of various areas of the physical examination center, and dynamic adjustment of the order of physical examination execution based on the priority weight of physical examination items; S52: Establish a real-time communication link between the smart terminal and the physical examination equipment to synchronously collect real-time vital sign data, compare the real-time vital sign data with a preset threshold to trigger a graded early warning mechanism, and synchronize the early warning information to the pre-bound caregiver terminal. S53: Collect user comfort feedback and urgent needs during the physical examination process through a pop-up questionnaire on a smart terminal, and quickly generate a process adjustment plan based on the feedback information; S6: Physical examination result comparison, correction and feedback iteration: Collect the current physical examination results and compare and correct them with the analysis results, and update the user profile and model parameters.

2. The method for health information push and physical examination optimization according to claim 1, characterized in that, In S1, a physical examination registration operation is performed and static basic information and health requests are supplemented. The static basic information includes the user's age, basic medical history, and occupation-related data. The health requests include the need for preparing for pregnancy, the need for chronic disease management, and the core purpose of the physical examination. The static basic information and health requests together constitute a basic information package.

3. The method for health information push and physical examination optimization according to claim 1, characterized in that, S2 includes the following sub-steps: S21: Integrate the static data in the basic information package with health needs to form a multi-dimensional user profile covering the user's health status, risk tendency and demand orientation; S22: Based on the multi-dimensional user profile, set weighting factors and calculate the priority weight of each physical examination item through the weighting allocation formula; S23: Connect to the historical database of physical examination users to extract similar user adaptation cases, remove extreme values ​​in the weight allocation results, and at the same time verify the matching degree between the weight allocation results and the equipment and doctor resources of the physical examination center to form the final personalized adaptation parameters including the weight of physical examination items and the priority of data collection.

4. The method for health information push and physical examination optimization according to claim 1, characterized in that, In step S3, multi-source data is acquired according to the adaptation parameters and standardized. The multi-source data includes historical physical examination data obtained from the physical examination user historical database, physical examination recommendation order information obtained from the physical examination intelligent order recommendation module, and monitoring data collected from mobile health APP and sports APP. The monitoring data includes heart rate monitoring data, sleep time data, and exercise-related data.

5. The method for health information push and physical examination optimization according to claim 1, characterized in that, In step S3, multi-source data is acquired according to the adaptation parameters and standardized. The standardization process includes cleaning the multi-source data using a preset standardization algorithm, filtering redundant data and outliers, and converting unstructured data into structured labels or quantitative indicators.

6. The method for health information push and physical examination optimization according to claim 1, characterized in that, S4 includes the following sub-steps: S41: Multimodal data fusion modeling adopts an attention mechanism to integrate the standardized structured data and unstructured data features from S3 to explore the potential correlations between data of different dimensions; S42: Based on the fused dataset in S41, the model parameters are updated by improving the incremental learning algorithm to perform real-time iterative optimization of the prediction model; S43: Based on the prediction model analysis results optimized from S42, the analysis results are classified according to the urgency of health risks and user awareness. High-risk items are given priority push permissions and are accompanied by a doctor intervention reminder icon, while routine items are accompanied by personalized optimization suggestions based on user profiles.

7. The method for health information push and physical examination optimization according to claim 1, characterized in that, The dynamic optimization of the physical examination order in S5 includes real-time collection of physical examination user's physical status data and physical examination center traffic data, and adjustment of the physical examination execution order based on the priority weight of the physical examination items. When the user is in a fatigued state, non-invasive rapid physical examination items are given priority.

8. The method for health information push and physical examination optimization according to claim 1, characterized in that, The health warning in S5 includes establishing a real-time communication link between the smart terminal and the physical examination equipment, synchronously collecting real-time vital sign data and comparing it with preset thresholds, triggering a graded warning mechanism, with different warning levels corresponding to different reminder methods and response measures.

9. The method for health information push and physical examination optimization according to claim 1, characterized in that, The S6 update of user profile and model parameters includes incorporating abnormal items in the current physical examination results into the user health risk label system, supplementing user data sensitivity parameters, and updating the parameters and weight factors of the prediction model based on the corrected deviation value.

10. The method for health information push and physical examination optimization according to claim 1, characterized in that, The personalized adaptation parameters and multi-source data acquisition form a linkage optimization. Monitoring data corresponding to high-weight physical examination items are collected first. The completeness and accuracy of the monitoring data are fed back to the adaptation parameter verification stage, triggering secondary verification of the adaptation parameters or adjustment of the weight of the corresponding physical examination items.

Images