Personalized digital health assessment method based on digital twinning and generative AI
By combining digital twins and generative AI technologies, multimodal data is collected to build models and generate health status assessment reports, which solves the problems of lack of personalization and accuracy in traditional health assessments and realizes personalized health assessments and personalized recommendations.
Patent Information
- Application Number
- CN202510688008.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-27
- Publication Date
- 2025-09-05
- Estimated Expiration
- Not applicable · inactive patent
Smart Images
Figure CN120600306A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent health management technology, and in particular to a personalized digital health assessment method based on digital twins and generative AI. Background Art
[0002] The core of digital twin technology is the interactive mapping of digital models and physical entities in virtual space. In recent years, digital twin technology has been widely applied in various fields, such as aerospace, engineering construction, and intelligent manufacturing. In healthcare, digital twin technology can be used to build a patient's "medical digital twin." By integrating information such as a patient's health records, medical history, and monitoring data from smart wearable devices, it simulates human functions in the cloud, enabling real-time monitoring of the patient's health status, predictive analysis, and precision medical diagnosis. Generative AI is a key branch of artificial intelligence. By learning from large amounts of data and patterns, it can generate new content such as text, images, and audio. In the field of health management, generative AI can be used to process complex multimodal data such as electrocardiograms, pulse waves, and facial spectra, enabling rapid assessment and prediction of health indicators.
[0003] With rising health awareness and limited medical resources, the demand for personalized digital health assessments is growing. Traditional health assessment methods often rely on physician experience and limited testing methods, making them unable to meet the demand for accurate, efficient, and personalized health assessments. While existing digital health assessment tools have improved efficiency and accuracy to a certain extent, they still lack personalization and precision. Therefore, a new approach is needed that can fully leverage the advantages of digital twins and generative AI technologies to achieve comprehensive, accurate, and real-time assessments of individual health status and provide personalized health recommendations and intervention plans. The development of digital twins and generative AI technologies offers new possibilities for personalized digital health assessments. By combining these technologies, it is hoped that the limitations of traditional health assessment methods can be overcome, providing people with more accurate, efficient, and personalized health assessment services. To this end, a personalized digital health assessment method based on digital twins and generative AI is proposed. Summary of the Invention
[0004] The main purpose of the present invention is to provide a personalized digital health assessment method based on digital twins and generative AI, which can effectively solve the problems in the background technology.
[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:
[0006] Personalized digital health assessment method based on digital twin and generative AI, including:
[0007] Step 1: Collect multimodal data related to the health of the user to be evaluated, where the multimodal data includes clinical information, real-time physiological data, lifestyle data, health records, medical history, medication history, and monitoring data obtained using smart wearable devices. Use data processing technology to integrate the obtained multi-source heterogeneous data to construct a health dataset for the user to be evaluated;
[0008] Step 2: Based on the integrated multimodal data, mathematical modeling and computer simulation technology are used to construct a digital twin model that reflects the health status of the user to be evaluated in real time. The digital twin model is also used to simulate the development of the target disease in the user to be evaluated;
[0009] Step 3: Use the generative AI model to learn the integrated multimodal data, obtain learning results including the data distribution status and the logical relationship between variables, and based on the data learning results, sample data from the data distribution and variable relationships of the constructed digital twin model to create synthetic data of multimodal data;
[0010] Step 4: Import the synthesized data of the multimodal data into the digital twin model to generate a current health status assessment report of the user to be assessed, as well as the risk and risk level of the target disease based on the current health status.
[0011] The method further comprises:
[0012] Step 5: Set the data update cycle to T, update the multimodal data in the health data set with the update cycle T, and dynamically update the digital twin model and the synthetic data of the multimodal data according to the health data set update result.
[0013] Furthermore, in step 2, the evaluation process of the health status of the user to be evaluated includes the following steps:
[0014] Construct a health status assessment indicator system and define the health status levels, where the health status levels are defined as healthy state, basic health state, sub-health state and unhealthy state;
[0015] Constructing a membership function for describing the degree of membership of any health status assessment indicator under different levels of health status, determining the health status assessment indicator data of the user in the current state based on the collected multimodal data of the user to be assessed, and performing preprocessing including data cleaning and standardization;
[0016] According to the constructed membership function and the obtained health status assessment index data, the membership of each evaluation index under different health status levels is calculated;
[0017] The membership of all evaluation indicators is weighted to obtain the health status evaluation result of the user to be evaluated, and the health status level of the user is determined based on the health status evaluation result.
[0018] Furthermore, the evaluation indicators include one or more combinations of physical health indicators, mental health indicators, social health indicators, lifestyle indicators, user behavior indicators, environmental factor indicators and self-assessment indicators.
[0019] Furthermore, in step 2, the specific process of using the digital twin model to simulate the development process of the target disease of the user to be evaluated includes the following steps:
[0020] Determining the type of data required for simulating the development process of the target disease, and extracting the required data from the collected multimodal data of the user to be evaluated based on the determination result;
[0021] Obtain historical actual case data of the development process of the target disease as learning corpus, use the learning corpus to train the digital twin model so that it can learn the development pattern and laws of the disease, and adjust the relevant parameters of the digital twin model until its accuracy and reliability meet the set expected values by comparing with the actual case data;
[0022] The current health status of the user to be evaluated is used as the initial condition, and the initial parameters of the digital twin model are set according to the initial conditions. By adjusting the parameters in the model, the progression of the disease under different conditions can be simulated;
[0023] Update the multimodal data in the health dataset, and dynamically update the digital twin model based on the health dataset update results, so that the accuracy and reliability of the digital twin model simulation results are not lower than the set expected values
[0024] Establish a feedback mechanism to continuously optimize the constructed digital twin model based on the actual health changes and feedback of the users to be evaluated.
[0025] Furthermore, the physical health indicator includes at least one of a physiological indicator, a disease state, a physical function, and a health risk behavior;
[0026] The mental health indicator includes at least one of emotional state, psychological resilience and cognitive function;
[0027] The social health indicator includes at least one of social support, social participation and social adaptability;
[0028] The lifestyle indicator includes at least one of eating habits, exercise habits and sleep quality;
[0029] The user behavior indicator includes at least one of preventive health care behavior and self-management behavior;
[0030] The environmental factor indicators include at least one of the living environment and the working environment;
[0031] The self-assessment indicator includes at least one of self-health assessment and quality of life assessment.
[0032] The present invention has the following beneficial effects:
[0033] Compared with the existing technology, by collecting multimodal data related to the health of the user to be evaluated, the obtained multi-source heterogeneous data is integrated using data processing technology to construct a health data set of the user to be evaluated. Based on the integrated multimodal data, mathematical modeling and computer simulation technology are used to construct a digital twin model for real-time reflection of the health status of the user to be evaluated. The digital twin model is also used to simulate the development process of the target disease of the user to be evaluated. The integrated multimodal data is learned using a generative AI model to obtain learning results including data distribution status and logical relationships between variables. Based on the data learning results, data sampling is performed from the data distribution and variable relationship of the constructed digital twin model to create synthetic data of multimodal data. The synthetic data of the multimodal data is imported into the digital twin model to generate a current health status assessment report of the user to be evaluated, as well as the risk and risk level of the target disease under the current health status. The advantages of digital twin and generative AI technologies can be fully utilized. By combining technologies, the limitations of traditional health assessment methods are overcome to achieve a comprehensive, accurate and real-time assessment of individual health status. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 This is a schematic diagram of the overall structure of the personalized digital health assessment method based on digital twins and generative AI of the present invention;
[0035] Figure 2 Schematic diagram of the evaluation process of the health status of the user to be evaluated in an embodiment of the solution of the present invention;
[0036] Figure 3 This is a schematic diagram of a simulation process of the target disease development process of a user to be evaluated in an embodiment of the solution of the present invention. DETAILED DESCRIPTION
[0037] The present invention will be further described below in conjunction with specific embodiments. The accompanying drawings are for illustrative purposes only and represent only schematic diagrams rather than actual drawings. They should not be understood as limiting the present invention. In order to better illustrate the specific embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product.
[0038] The specific implementation process of the technical solution of the present invention includes the following steps:
[0039] Step 1: Collect multimodal data related to the health of the user to be evaluated. Multimodal data includes clinical information, real-time physiological data, lifestyle data, health records, medical history, medication history, and monitoring data obtained using smart wearable devices. Use data processing technology to integrate the obtained multi-source heterogeneous data to construct a health dataset for the user to be evaluated. The specific process of multimodal data integration can be divided into the following key steps, which cover the entire process from data collection to fusion model design and optimization:
[0040] Step S11: Data collection and preprocessing
[0041] Data collection: Collect data of different modalities from multiple channels, such as images, text, audio, sensor data, etc.
[0042] Data cleaning: remove noise, fill missing values, unify timestamps, and other operations to ensure data quality.
[0043] Standardization processing: Convert data in different formats into a unified standard format to facilitate subsequent processing.
[0044] Step S12: Feature extraction
[0045] Image feature extraction: Use methods such as convolutional neural networks (CNN) to process image data.
[0046] Text feature extraction: Use pre-trained language models (such as BERT, GPT) or word embedding methods to process text data.
[0047] Other modal feature extraction: Select appropriate feature extraction methods based on the data type, such as long short-term memory network (LSTM) to process sequence data.
[0048] Step S13: Data alignment
[0049] Temporal and spatial alignment: Data from different sources are temporally and spatially aligned to ensure consistency of the fused data.
[0050] Step S14: Data fusion
[0051] Early Fusion: Directly fuse raw data or low-level features from different modalities at the input layer. For example, concatenating image pixels and text word vectors before inputting them into the model.
[0052] Middle Fusion (Representation-level Fusion): Each modality is first encoded separately, and then interacted through an attention mechanism or graph network. For example, cross-modal attention in vision-language models (such as CLIP)
[0053] Late Fusion: Fusion the scores (decisions) output by classifiers trained on different modal data.
[0054] Step S15: Fusion model design and optimization
[0055] Select a fusion model: Based on the application scenario and data characteristics, select an appropriate fusion model, such as weighted average, neural network, support vector machine, etc.
[0056] Optimize fusion strategy: Use methods such as shared attention mechanism to reduce computational complexity and improve the model's adaptability to weak modality data.
[0057] Step S16: Application and Evaluation
[0058] Application scenario development: Apply the integrated data to specific tasks, such as autonomous driving, medical diagnosis, sentiment analysis, etc.
[0059] Dataset evaluation: Evaluate the quality, diversity, representativeness, and other aspects of the dataset to ensure that it meets actual needs.
[0060] Through the above process, multimodal data can be effectively integrated to provide richer information and more accurate decision support for various application scenarios.
[0061] Step 2: Based on the integrated multimodal data, mathematical modeling and computer simulation technology are used to construct a digital twin model that reflects the health status of the user to be evaluated in real time. The digital twin model is also used to simulate the development of the target disease in the user to be evaluated;
[0062] The evaluation process of the health status of the user to be evaluated includes the following steps:
[0063] Step S211: constructing a health status assessment indicator system and defining health status levels, wherein the health status levels are sequentially defined into four categories: healthy state, basic healthy state, sub-healthy state, and unhealthy state;
[0064] Step S212: Constructing a membership function for describing the degree of membership of any health status assessment indicator under different health status levels, determining the health status assessment indicator data of the user in the current state based on the collected multimodal data of the user to be assessed, and performing preprocessing including data cleaning and standardization;
[0065] Step S213: Calculating the membership of each evaluation indicator at different health status levels based on the constructed membership function and the acquired health status evaluation index data;
[0066] Step S214: weighting the membership of all evaluation indicators to obtain the health status evaluation result of the user to be evaluated, and determining the health status level of the user according to the health status evaluation result.
[0067] The specific process of simulating the development of the target disease for the user to be evaluated includes the following steps:
[0068] Step S221: determining the data type required for simulating the development process of the target disease, and extracting the required data from the collected multimodal data of the user to be evaluated based on the determination result;
[0069] Step S222: Acquire historical actual case data of the development process of the target disease as learning corpus, use the learning corpus to train the digital twin model so that it can learn the development pattern and laws of the disease, and adjust the relevant parameters of the digital twin model until its accuracy and reliability meet the set expected values by comparing with the actual case data;
[0070] Step S223: Using the current health status of the user to be evaluated as the initial condition, the initial parameters of the digital twin model are set according to the initial conditions, and the parameters in the model are adjusted to simulate the progression of the disease under different conditions;
[0071] Step S224: updating the multimodal data in the health dataset, and dynamically updating the digital twin model based on the health dataset update results, so that the accuracy and reliability of the digital twin model simulation results are not lower than the set expected values;
[0072] Step S225: Establish a feedback mechanism to continuously optimize the constructed digital twin model based on the actual health changes and feedback of the user to be evaluated.
[0073] It should be noted that health status assessment indicators usually cover multiple dimensions to comprehensively reflect an individual's health status. The following are some common health status assessment indicators, referring to the latest research and official guidelines:
[0074] Physical health
[0075] Physiological indicators: including blood pressure, blood sugar level, blood lipid level (such as total cholesterol, HDL cholesterol, LDL cholesterol, blood triglycerides), heart rate, etc.
[0076] Disease status: such as cancer stage and control of chronic diseases (such as diabetes and cardiovascular disease).
[0077] Physical function: such as body mass index (BMI), physical activity level, muscle strength, joint flexibility, etc.
[0078] Health risk behaviors: such as smoking, drinking, drug abuse, etc.
[0079] Mental Health
[0080] Emotional state: such as depression, anxiety, mood swings, etc.
[0081] Psychological resilience: such as the ability to cope with stress, life satisfaction, etc.
[0082] Cognitive functions: such as memory, attention, and mental agility.
[0083] Social Health
[0084] Social support: such as family relationships, social networks, and the quality of social support systems.
[0085] Social participation: such as whether to participate in community activities, volunteer activities, etc.
[0086] Social adaptability: such as job satisfaction, life adaptability, and fulfillment of social roles.
[0087] lifestyle
[0088] Eating habits: such as nutritional balance, whether eating regularly, and whether there are any specific dietary restrictions.
[0089] Exercise habits: such as weekly exercise frequency, exercise type, exercise intensity, etc.
[0090] Sleep quality: such as sleep duration, sleep quality, and sleep disorders.
[0091] User Conduct
[0092] Preventive health care behaviors: such as regular physical examinations, vaccinations, cancer screening, etc.
[0093] Self-management behavior: such as the ability of chronic disease patients to manage their own diseases and medication compliance.
[0094] Environmental factors
[0095] Living environment: such as air quality, safety of living environment, noise level, etc.
[0096] Working environment: such as work pressure, occupational exposure risks, etc.
[0097] self assessment
[0098] Self-health evaluation: such as an individual's overall evaluation of his or her health status.
[0099] Quality of life assessment: such as life satisfaction, degree of activity limitation, etc.
[0100] These indicators can be adjusted and selected according to different evaluation purposes and objects to ensure the accuracy and comprehensiveness of the evaluation results.
[0101] Step 3: Use the generative AI model to learn the integrated multimodal data and obtain learning results including the data distribution status and the logical relationship between variables. Based on the data learning results, data is sampled from the data distribution and variable relationships of the constructed digital twin model to create synthetic data of the multimodal data. The synthetic data can be generated through the following process, specifically the following steps:
[0102] Step S31: Data understanding and preprocessing
[0103] Data understanding: First, fully understand the real data set, including the distribution of the data, the relationship between variables, missing data elements and extreme values.
[0104] Data preprocessing: Remove or fill in missing values, correct errors, and standardize the data format. Also, remove or encrypt any personally identifiable information (PII) to ensure privacy.
[0105] Step S32: Selecting synthetic data generation technology
[0106] Deep Learning Models:
[0107] Generative Adversarial Networks (GANs): GANs consist of a generator and a discriminator. The generator creates synthetic data, while the discriminator distinguishes between synthetic data and real data. Through adversarial training between the two, the generator can generate synthetic data with a distribution similar to real data.
[0108] Variational Autoencoder (VAE): A VAE is an unsupervised machine learning model in which an encoder compresses and integrates real-world data, while a decoder analyzes this data to generate a representation of the real-world data. The key advantage of a VAE is that it ensures that the input and output data remain highly similar.
[0109] Generative Pre-Trained Transformer (GPT): GPT is a generative model based on the Transformer architecture that can generate high-quality text data. In the medical field, it can be used to generate text descriptions related to patient health.
[0110] Data augmentation: Although data augmentation is not synthetic data, it can improve the generalization ability of the model by adding new data to the existing dataset.
[0111] Step S33: Synthetic data creation and verification
[0112] Synthetic data creation: Creating new data points by sampling from modeled distributions and relationships, ensuring that the synthetic data retains the statistical properties of the original data.
[0113] Post-processing: Alters synthetic data by adding subtle nuances and avoiding consistent patterns that suggest artificial creation. Inspects synthetic data to confirm it does not contain any re-identifiable personal information.
[0114] Validation and practical evaluation: Validate the quality of synthetic data to ensure it is statistically similar to real data, and evaluate its effectiveness in real-world applications.
[0115] Step S34: Applying synthetic data to optimize the digital twin model
[0116] Filling data gaps: Synthetic data can generate data and long-tail data in some sensitive or high-security areas, thereby filling gaps in real data and improving the accuracy of the twin model.
[0117] Enhanced model training: Using synthetic data for model training can improve the generalization and adaptability of the model, enabling it to better simulate the patient's health status and the progression of the target disease.
[0118] Dynamic Updates and Optimization: As new data is generated and models are trained, the digital twin model is continuously updated to achieve more accurate health predictions and personalized intervention plans.
[0119] Through the above methods, generative AI can generate synthetic data for digital twin models used to optimize patient health, providing strong support for research and practice in the medical field.
[0120] Step 4: Import the synthesized multimodal data into the digital twin model to generate a report on the user's current health status, as well as the target disease's risk and risk level based on the user's current health status. The method for determining the risk and risk level is described in steps S211-S214 and will not be repeated here.
[0121] Step 5: Set the data update cycle to T, update the multimodal data in the health dataset with the update cycle T, and dynamically update the digital twin model and the synthetic data of the multimodal data based on the update results of the health dataset.
[0122] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.
Claims
1. A personalized digital health assessment method based on digital twins and generative AI, characterized by: include: Step 1: Collect multimodal data related to the health of the user to be evaluated, where the multimodal data includes clinical information, real-time physiological data, lifestyle data, health records, medical history, medication history, and monitoring data obtained using smart wearable devices. Use data processing technology to integrate the obtained multi-source heterogeneous data to construct a health dataset for the user to be evaluated; Step 2: Based on the integrated multimodal data, mathematical modeling and computer simulation technology are used to construct a digital twin model that reflects the health status of the user to be evaluated in real time. The digital twin model is also used to simulate the development of the target disease in the user to be evaluated; Step 3: Use the generative AI model to learn the integrated multimodal data, obtain learning results including the data distribution status and the logical relationship between variables, and based on the data learning results, sample data from the data distribution and variable relationships of the constructed digital twin model to create synthetic data of multimodal data; Step 4: Import the synthesized data of the multimodal data into the digital twin model to generate a current health status assessment report of the user to be assessed, as well as the risk and risk level of the target disease based on the current health status.
2. The personalized digital health assessment method based on digital twins and generative AI according to claim 1, characterized in that The method further comprises: Step 5: Set the data update cycle to T, update the multimodal data in the health data set with the update cycle T, and dynamically update the digital twin model and the synthetic data of the multimodal data according to the health data set update result.
3. The personalized digital health assessment method based on digital twins and generative AI according to claim 1, characterized in that In step 2, the evaluation process of the user's health status to be evaluated includes the following steps: Construct a health status assessment indicator system and define the health status levels, where the health status levels are defined as healthy state, basic health state, sub-health state and unhealthy state; Constructing a membership function for describing the degree of membership of any health status assessment indicator under different levels of health status, determining the health status assessment indicator data of the user in the current state based on the collected multimodal data of the user to be assessed, and performing preprocessing including data cleaning and standardization; According to the constructed membership function and the obtained health status assessment index data, the membership of each evaluation index under different health status levels is calculated; The membership of all evaluation indicators is weighted to obtain the health status evaluation result of the user to be evaluated, and the health status level of the user is determined based on the health status evaluation result.
4. The personalized digital health assessment method based on digital twins and generative AI according to claim 1, characterized in that The evaluation indicators include one or more combinations of physical health indicators, mental health indicators, social health indicators, lifestyle indicators, user behavior indicators, environmental factor indicators and self-assessment indicators.
5. The personalized digital health assessment method based on digital twins and generative AI according to claim 1, characterized in that In step 2, the specific process of using the digital twin model to simulate the development of the target disease for the user to be evaluated includes the following steps: Determining the type of data required for simulating the development process of the target disease, and extracting the required data from the collected multimodal data of the user to be evaluated based on the determination result; Obtain historical actual case data of the development process of the target disease as learning corpus, use the learning corpus to train the digital twin model so that it can learn the development pattern and laws of the disease, and adjust the relevant parameters of the digital twin model until its accuracy and reliability meet the set expected values by comparing with the actual case data; The current health status of the user to be evaluated is used as the initial condition, and the initial parameters of the digital twin model are set according to the initial conditions. By adjusting the parameters in the model, the progression of the disease under different conditions can be simulated; Update the multimodal data in the health dataset, and dynamically update the digital twin model based on the health dataset update results, so that the accuracy and reliability of the digital twin model simulation results are not lower than the set expected values Establish a feedback mechanism to continuously optimize the constructed digital twin model based on the actual health changes and feedback of the users to be evaluated.
6. The personalized digital health assessment method based on digital twins and generative AI according to claim 4, characterized in that The physical health indicator includes at least one of a physiological indicator, a disease state, a physical function, and a health risk behavior; The mental health indicator includes at least one of emotional state, psychological resilience and cognitive function; The social health indicator includes at least one of social support, social participation and social adaptability; The lifestyle indicator includes at least one of eating habits, exercise habits and sleep quality; The user behavior indicator includes at least one of preventive health care behavior and self-management behavior; The environmental factor indicators include at least one of the living environment and the working environment; The self-assessment indicator includes at least one of self-health assessment and quality of life assessment.