Health management method and device based on artificial intelligence, equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a health management method and device based on artificial intelligence, equipment and a storage medium, and the method comprises the steps: obtaining a multi-modal fusion health feature of a user; constructing a health knowledge graph of the user based on the multi-modal fusion health features; retrieving knowledge sub-graphs related to the multi-modal fusion health features from a health knowledge graph; performing prediction processing on the health risk of the user according to the multi-modal fusion health features and the knowledge sub-graph through a preset medical large model to obtain a health risk probability of the user; and through a reinforcement learning algorithm, health intervention measures are generated for the user according to the multi-modal fusion health features, the knowledge sub-graph and the health risk probability. The method can be applied to an individual health management scene in the field of medical science and technology, and can improve the accuracy, effectiveness and effect of individual health management.

Description

AI-based health management methods, devices, equipment, and storage media Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a health management method, apparatus, device, and storage medium based on artificial intelligence. Background Technology

[0002] In the field of medical technology, with the increasing public awareness of health and technological advancements, health monitoring devices such as smartwatches and smart bracelets are being increasingly widely used in daily life. These devices, utilizing advanced sensing technology, can collect various health indicators in real time, such as heart rate, blood oxygen saturation, and sleep quality, helping users manage their health in real time.

[0003] Despite significant technological advancements in health monitoring devices, some key shortcomings remain in data management and application effectiveness. Specifically, the health indicators collected by these devices are often isolated and fragmented, making it difficult for users to gain comprehensive health insights and impacting the accuracy and effectiveness of individual health management.

[0004] Secondly, although health monitoring devices have the function of health risk reminders, they are limited to passive monitoring and recording, lacking the necessary foresight (i.e., unable to deduce potential health trends) and proactive intervention capabilities. This means that users are unlikely to receive timely guidance and support when facing health risks, affecting the effectiveness of individual health management. Summary of the Invention

[0005] This invention provides an artificial intelligence-based health management method, device, equipment, and storage medium, aiming to address the technical problem that existing health management methods fail to provide users with comprehensive health insights and lack necessary predictive and proactive intervention capabilities, thus affecting the accuracy, effectiveness, and impact of individual health management. In a first aspect, an artificial intelligence-based health management method is provided, comprising: acquiring a user's multimodal fusion health characteristics; constructing a health knowledge graph of the user based on the multimodal fusion health characteristics; retrieving knowledge subgraphs related to the multimodal fusion health characteristics from the health knowledge graph; predicting the user's health risks based on the multimodal fusion health characteristics and the knowledge subgraphs using a pre-set medical large-scale model to obtain the user's health risk probability; and generating health intervention measures for the user based on the multimodal fusion health characteristics, the knowledge subgraphs, and the health risk probability using a reinforcement learning algorithm.

[0006] Secondly, an artificial intelligence-based health management device is provided, comprising: a feature acquisition module for acquiring a user's multimodal fused health features; a graph construction module for constructing a health knowledge graph of the user based on the multimodal fused health features; a subgraph retrieval module for retrieving knowledge subgraphs related to the multimodal fused health features from the health knowledge graph; a risk prediction module for predicting the user's health risk based on the multimodal fused health features and the knowledge subgraph using a preset medical big data model, thereby obtaining the user's health risk probability; and a measure generation module for generating health intervention measures for the user based on the multimodal fused health features, the knowledge subgraph, and the health risk probability using a reinforcement learning algorithm.

[0007] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described artificial intelligence-based health management method.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned artificial intelligence-based health management method.

[0009] The aforementioned AI-based health management methods, devices, equipment, and storage media achieve two main goals. First, by acquiring multimodal fusion health features that comprehensively reflect a user's overall health status, a powerful, flexible health knowledge graph encompassing the causal relationships between relevant health events under the user's overall health status is constructed, providing theoretical support and data basis for health management. Second, knowledge subgraphs related to the multimodal fusion health features are retrieved from the health knowledge graph. This allows for efficient and intelligent prediction of the user's health risks using a large-scale medical model based on the multimodal fusion health features and knowledge subgraphs, thus obtaining the user's health... The concept of health risk probability allows large-scale medical models to comprehensively consider the overall health status reflected by multimodal fusion health features and the key medical knowledge provided by knowledge subgraphs when making risk predictions, thereby improving the accuracy of health risk probability and enhancing the predictability of health management. On the other hand, it abstracts health interventions for users into reinforcement learning tasks. Using reinforcement learning algorithms, personalized health intervention measures are generated for users based on multimodal fusion health features, knowledge subgraphs, and health risk probabilities. This will provide users with accurate and scientific guidance for health management, enhance the proactive intervention capability of health management, and significantly improve the accuracy, effectiveness, efficiency, and impact of health management. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 is a flowchart of a health management method based on artificial intelligence in one embodiment of the present invention; Figure 2 is a flowchart of a specific implementation of step S20 in Figure 1; Figure 3 is a flowchart of a specific implementation of step S50 in Figure 1; Figure 4 is a structural schematic diagram of a health management device based on artificial intelligence in one embodiment of the present invention; Figure 5 is a structural schematic diagram of a computer device in one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The AI-based health management method provided in this invention can be applied to a server. The server can acquire a user's multimodal fusion health characteristics; construct a user's health knowledge graph based on the multimodal fusion health characteristics; retrieve knowledge subgraphs related to the multimodal fusion health characteristics from the health knowledge graph; predict the user's health risks based on the multimodal fusion health characteristics and knowledge subgraphs using a pre-set medical big data model to obtain the user's health risk probability; and generate health intervention measures for the user based on the multimodal fusion health characteristics, knowledge subgraphs, and health risk probability using a reinforcement learning algorithm. In this way, on the one hand, by acquiring multimodal fusion health features that comprehensively reflect the user's overall health status, a powerful, flexible health knowledge graph covering the causal relationships between relevant health events under the user's overall health status can be constructed, providing theoretical support and data basis for health management. On the other hand, knowledge subgraphs related to multimodal fusion health features can be retrieved from the health knowledge graph, enabling efficient and intelligent prediction of the user's health risks through a large medical model based on multimodal fusion health features and knowledge subgraphs, obtaining the user's health risk probability. This allows the large medical model to comprehensively consider the overall health status reflected by multimodal fusion health features and the key medical knowledge provided by knowledge subgraphs when making risk predictions, improving the accuracy of health risk probabilities and enhancing the predictability of health management. Furthermore, by abstracting user health interventions into reinforcement learning tasks, reinforcement learning algorithms can be used to generate personalized health intervention measures for users based on multimodal fusion health features, knowledge subgraphs, and health risk probabilities. This will provide accurate and scientific guidance for user health management, enhance the proactive intervention capability of health management, and significantly improve the accuracy, effectiveness, efficiency, and impact of health management. The server-side can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0014] The artificial intelligence-based health management method provided by this invention can be applied to individual health management scenarios in the field of medical technology. Through a closed-loop health management mechanism incorporating multiple artificial intelligence algorithms, it achieves a closed-loop health management process from "passive monitoring" to "active prediction" and then to "personalized intervention," ensuring that users receive timely, effective, accurate, and personalized health support tailored to their individual needs.

[0015] To facilitate understanding, the terms involved in this invention will first be explained: Artificial intelligence (AI): is a new technical science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. AI also refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0016] Med-LLM (Medical Large Language Model) refers to a large language model (LLM) specifically designed for the medical and healthcare fields, trained on large-scale medical corpora. It combines Natural Language Processing (NLP) techniques with medical knowledge to understand, generate, and process natural language information in the medical and healthcare domain, playing a role in various medical and healthcare-related tasks. Examples of medical large models include PubMedGPT, GatorTron, Med-PaLM 2, HuatuoGPT, MedGPT, PMC-LLaMA, etc., without limitation. It should be noted that in this embodiment of the invention, the medical large model is pre-tuned through medical knowledge distillation and RLHF (Reinforcement Learning with Human Feedback) to enable it to output results with professional reasoning chains, explanations, and human comprehensibility in medical tasks, rather than simply providing a conclusion.

[0017] Medical knowledge distillation: By integrating medical expertise into a large-scale medical model, the model can not only provide accurate health risk predictions but also offer professional medical explanations.

[0018] RLHF Fine-tuning: By incorporating human feedback for fine-tuning, large medical models can generate more interpretable and credible health risk predictions.

[0019] Transformer model: It is a deep learning model architecture based on attention mechanism, with powerful parallel computing capabilities, and can effectively integrate features from different modalities or sources.

[0020] Reinforcement Learning (RL) is a branch of machine learning that focuses on learning optimal policies through interaction with the environment to maximize rewards. In reinforcement learning, an agent interacts with the environment by taking a series of actions and adjusts its policy based on the feedback received (rewards or penalties). This process enables the agent to optimize its behavior in a specific context to achieve a predetermined goal.

[0021] Based on this, the health management method provided by the present invention will be described in detail below.

[0022] Please refer to Figure 1, which is a flowchart of an artificial intelligence-based health management method provided in an embodiment of the present invention, including the following steps: S10: Obtain the user's multimodal fusion health characteristics.

[0023] For step S10, the user's multimodal fusion health features can be obtained, which can reflect the user's overall health status.

[0024] In step S10 of some embodiments, the user's multimodal health features can be obtained; the multimodal health features are fused using a preset Transformer model to obtain multimodal fused health features.

[0025] First, multimodal health data of users can be collected. Multimodal health data can include physiological data, behavioral data, environmental data, and emotional data, etc.

[0026] Physiological data such as heart rate (HR), blood oxygen saturation (SpO2), skin conductance (EDA), and body temperature (T) are recorded as follows: .

[0027] Behavioral data, such as steps, posture angles, acceleration, and other motion states, are denoted as... .

[0028] Environmental data, such as temperature, humidity, air pressure, and air quality, are denoted as... .

[0029] Emotional data, such as speech signal features like intensity, rate of speech, and rhythm, are denoted as... And facial expression recognition information or emotional state (such as joy, anger, sadness, anxiety, etc.), recorded as .

[0030] Multimodal health data of users can be stably collected through various sensors (such as wearable devices, home sensors, or mobile apps), denoted as a matrix:

[0031] Among them, matrix Includes all modal health data at time step t.

[0032] To ensure the temporal consistency, reliability, and availability of multimodal health data, preprocessing can be performed to obtain multimodal health features, denoted as... This provides rich foundational features for subsequent health knowledge graph construction, health risk prediction, and health intervention decision-making.

[0033] The preprocessing includes time alignment and outlier correction.

[0034] Specifically, considering that multimodal health data is collected through different sensors and may have different sampling frequencies or timestamps, time alignment processing can be performed on the multimodal health data to obtain time-aligned multimodal health data, ensuring the temporal consistency of the multimodal health data. For example, linear interpolation can be used to perform time alignment processing on multimodal health data, that is, for missing or misaligned data points, interpolation is performed using the values ​​of known neighboring points.

[0035] Considering that multimodal health data acquisition may be subject to interference (such as sensor failure or environmental interference), which may lead to outliers or noise in the multimodal health data, outlier detection methods (such as Z-score method) can be used to identify outliers in the time-aligned multimodal health data first. Then, the identified outliers can be corrected, for example, by using median filtering or mean correction methods, to obtain multimodal health features and improve the quality of the multimodal health features.

[0036] In this way, by performing time alignment and outlier correction preprocessing on multimodal health data, we can obtain time-consistent and reliable multimodal health features, which are convenient for subsequent use.

[0037] After obtaining the user's multimodal health features, the multimodal health features can be fused using a pre-defined Transformer model. This process captures the temporal dependencies and interactions between modalities within the multimodal health features, resulting in multimodal fused health features and enabling a unified representation of the user's multimodal health data.

[0038] In the Transformer model, the processing of multimodal health features can be described by the following formula:

[0039] in, Let be the modality mapping matrix, representing the mapping of multimodal health features. Mapped to a unified dimension This ensures that health characteristics of different modalities can be fused in the same space; The hidden state representation output by the Transformer model, also known as the multimodal fusion health feature, is used to describe the overall health state at time step t, and contains the fusion information of all modes at time t.

[0040] Specifically, the self-attention mechanism of the Transformer model can be used to calculate the relationships between each time step and between modes to capture temporal dependencies and interactions between modes. The self-attention calculation formula is as follows:

[0041] in, These are the query, key, and value, representing features from different time steps and modalities, respectively. Indicates calculation and The dot product is obtained and Similarity; Used for scaling and The similarity is used to prevent the inner product from becoming too large in high-dimensional space, which would lead to gradient vanishing.

[0042] In other words, in the attention mechanism, it is calculated through dot product. and The similarity is calculated, scaled, and then normalized using the softmax function to obtain a weight matrix. This weight matrix is ​​then compared with... Multiply the results and obtain a new representation by weighted averaging. This new representation can aggregate the dependency information between time and modality.

[0043] Through its self-attention mechanism, the Transformer model can not only effectively capture the relationships between different time steps in a time series, but also handle the information interaction between different modalities. For example, it can capture the correlation between physiological signals and emotional states, as well as the interaction patterns between speech signal features and motion states, thereby outputting multimodal fused health features. Where T represents the length of the time step (i.e., the time series length of the multimodal health features). The hidden dimension represents the feature dimension at each time step after processing by the Transformer model.

[0044] In this way, by efficiently fusing users' multimodal health features in the time and modal dimensions through the Transformer model, the output of multimodal fused health features that can reflect the user's overall health status can provide strong feature support for subsequent health knowledge graph construction, health risk prediction and health intervention decision-making.

[0045] S20: Construct a user's health knowledge graph based on multimodal fusion of health features.

[0046] In step S20, the user's multimodal health characteristics can be combined with knowledge in the medical field to construct a powerful and flexible structured knowledge tool—a health knowledge graph—to intuitively describe the causal relationships between relevant health events in the user's overall health status, providing theoretical support and data basis for health management.

[0047] In some embodiments, please refer to Figure 2. Step S20 may include, but is not limited to, the following steps: S21: Construct a health knowledge graph structure for the user; S22: Update the health knowledge graph structure based on multimodal fusion health features to obtain a health knowledge graph.

[0048] For steps S21-S22, a user's health knowledge graph structure can be constructed, denoted as... , At time t, it is defined as a triple:

[0049] in, A set of nodes represents health-related concepts or events, such as {"hypertension", "insufficient sleep", "excessive caffeine intake", "elevated heart rate"}; each node in the set represents a health state or health factor.

[0050] Let be a set of edges, representing semantic, cooperative, or causal relationships between different nodes (different health events). For example, "hypertension" may be connected to "insufficient sleep," indicating that "insufficient sleep" may lead to "hypertension."

[0051] This is a set of edge weights and node weights, reflecting the strength of the relationship and the importance of the nodes. Edge weights can represent the degree of correlation or causal influence between health events, while node weights reflect the importance of each health event at time t.

[0052] The health knowledge graph structure can be updated based on multimodal fusion of health features to obtain the user's health knowledge graph, so that the health knowledge graph is consistent with the user's health status.

[0053] In step S22 of some embodiments, a health risk score function can be calculated based on multimodal fusion health features; the node weights of the health knowledge graph structure can be updated based on the health risk score function; and the edge weights of the health knowledge graph structure can be updated through a preset medical knowledge base to obtain a health knowledge graph.

[0054] Specifically, the mechanism for updating the health knowledge graph structure is to update the node weights based on multimodal fusion of health features and to update the edge weights of the health knowledge graph structure based on a pre-set medical knowledge base.

[0055] (a) Node weight update: The weight of each node can be dynamically updated using the node weight update formula, which is shown below:

[0056] in, For time t node The weight of the node represents the node's weight. The current importance or degree of importance; For time t-1 node The weight of the node represents the node's weight. The weight at the previous moment; The time decay factor, ranging from 0.7 to 0.9, is used to balance the importance of historical information and current new data. This ensures that the health knowledge graph not only helps to reflect the user's health status in real time, but also maintains a certain historical context, thereby enhancing the interpretability and continuity of the health knowledge graph. It is a health risk score function calculated based on multimodal fusion of health features, reflecting the potential risks of the current health status.

[0057] This health risk score function can be calculated using a linear layer and a sigmoid activation function:

[0058] in, This represents the weight matrix of the linear layer; Indicates the bias term; This represents the Sigmoid activation function.

[0059] In other words, the health risk score function describes the mapping process from multimodal health features to risk scores: first, a linear transformation is performed on the multimodal fused health features (i.e., the multimodal fused health features are transformed). and (perform multiplication), so that It is linearly transformed into a new linear combination In the new linear combination Add bias term The result after linear transformation is obtained. Finally, the health risk score function is obtained by performing a non-linear mapping on the result of the linear transformation using an activation function. .

[0060] The health risk scoring function can dynamically reflect the latest changes in a user's overall health status, thereby supporting the updating of node weights in the health knowledge graph structure.

[0061] (ii) Edge weight update: edge weight express and The strength of the relationship between them.

[0062] Pre-defined medical knowledge bases, such as professional medical knowledge bases like PubMed and ICD-10 classification, can be used to supplement and validate edge weights, enabling edge weight updates and further enhancing the interpretability and medical evidence of the health knowledge graph. For example, if medical literature in the knowledge base indicates that "excessive caffeine intake" directly leads to "increased heart rate," then the edge weight between the two can be further increased based on evidence from the medical literature.

[0063] In some embodiments, topic co-occurrence analysis can also be used to update edge weights, that is, to dynamically adjust edge weights by analyzing the co-occurrence frequency of health events. For example, if "hypertension" and "insufficient sleep" co-occur in multiple health records, then the edge weight between them can be increased.

[0064] In this way, by constructing a health knowledge graph, we can not only intuitively represent the correlation between health events, but also support long-term health risk prediction and health intervention decision-making.

[0065] S30: Retrieve knowledge subgraphs related to multimodal fusion health features from the health knowledge graph.

[0066] For step S30, considering that the health knowledge graph may be very large, containing many nodes and edges, the RAG (Retrieval-Augmented Generation) mechanism can be used to retrieve and fuse health features with multimodal data. More relevant knowledge subgraphs A knowledge subgraph can be understood as a small portion of a graph structure that is semantically similar to the multimodal fused health features. The knowledge subgraph contains elements that are semantically similar to the multimodal fused health features. More relevant semantic information can provide more targeted medical causal relationships.

[0067] Knowledge subgraph retrieval via RAG mechanism The method is as follows:

[0068] in, As a semantic similarity function, cosine similarity or attention relevance can be used to measure the multimodal fusion health features at the current time t. With the set of nodes in the health knowledge graph semantic similarity; This represents the top K knowledge subgraphs with the highest semantic similarity, where K is a positive integer greater than or equal to 1. The specific value can be flexibly set according to actual needs and is not limited here.

[0069] In this way, based on semantic similarity, the K knowledge subgraphs most relevant to the multimodal fusion health features are retrieved. These knowledge subgraphs contain the most important nodes and relationships related to health events, providing crucial medical background information or knowledge for subsequent health risk prediction and health intervention generation.

[0070] S40: Using a pre-set medical big data model, the system predicts and processes the user's health risks based on multimodal fusion of health features and knowledge subgraphs to obtain the user's health risk probability.

[0071] For step S40, the user's health risk is efficiently and intelligently predicted using a large medical model based on multimodal fusion of health features and knowledge subgraphs, thereby obtaining the user's health risk probability and improving the efficiency and accuracy of health risk prediction.

[0072] In step S40 of some embodiments, the multimodal fusion features and knowledge subgraphs can be spliced ​​together to obtain a health status representation of the user; the health status representation is then input into a large medical model for health risk prediction to obtain the health risk probability.

[0073] Specifically, multimodal fusion features and knowledge subgraphs can be concatenated to obtain a user's health status representation, denoted as... As input to the medical big data model, it enables the model to comprehensively consider overall health status and medical knowledge, thereby providing more accurate risk predictions.

[0074] The concatenated health status representation is input into a large-scale medical model for health risk prediction, yielding the model's output:

[0075] in, This represents the output of the large medical model. include and Two parts; The probability of health risk refers to the probability of a certain health event (such as "abnormal blood sugar") occurring. Interpretable text indicating the probability of health risks, such as "recent high sugar intake + lack of exercise" causing "blood sugar fluctuations".

[0076] In other words, large-scale medical models can not only predict the probability of health risks, but also provide detailed and professional medical explanations of those probabilities.

[0077] S50: Through reinforcement learning algorithms, health intervention measures are generated for users based on multimodal fusion of health features, knowledge subgraphs, and health risk probabilities.

[0078] In step S50, the health intervention for users is abstracted into a reinforcement learning task. Through reinforcement learning algorithms, personalized health intervention measures are generated for users based on multimodal fusion of health features, knowledge subgraphs, and health risk probabilities. This provides accurate and scientific guidance for users' health management, thereby significantly improving the efficiency and effectiveness of health management.

[0079] In some embodiments, please refer to Figure 3. Step S50 may include, but is not limited to, the following steps: S51: Integrating multimodal fusion health features, knowledge subgraphs and health risk probabilities into the user's health risk status; S52: Calculating the rewards of the candidate intervention measures that can be selected under the health risk status to obtain the rewards of the candidate intervention measures; S53: Selecting the candidate intervention measure with the highest reward as the health intervention measure.

[0080] Specifically, user health interventions are abstracted as state-action tasks in reinforcement learning. Reinforcement learning algorithms are applied to learn how to select the most effective interventions under different health states, adapting to individual user differences to maximize user health improvement and satisfaction. Here, "State" refers to the health risk state, denoted as […]. Action: Represents available intervention options, including but not limited to medication reminders, dietary recommendations, exercise programs, and psychological interventions, etc., denoted as... Reward Function: The core of reinforcement learning lies in evaluating the effectiveness of each intervention through a reward function. In other words, the reward function is used to evaluate the effectiveness of selected interventions under a specific health risk state, and is defined as follows:

[0081] in, This represents the change in a user's overall health score, indicating the degree of health improvement brought about by the intervention. This indicates user satisfaction feedback via the mobile app, directly reflecting users' acceptance or satisfaction with the intervention measures; This indicates the degree to which the user deviates from medical advice during the implementation of intervention measures; They are respectively , and The weighting coefficients are used to balance their importance.

[0082] The optimization goal of the reward function is to select the most appropriate intervention by maximizing the change in the overall health score and user satisfaction, while minimizing the degree of deviation from medical advice.

[0083] Based on this, and using a predefined strategy generation formula, health intervention measures for users are generated. The strategy generation formula is as follows:

[0084] In other words, under specific health risk conditions The following approach will consider all possible candidate interventions, calculate the reward for each candidate intervention using a reward function, and select the candidate intervention that maximizes the reward (highest reward) as the health intervention.

[0085] After step S50 in some embodiments, real feedback information from users after taking health intervention measures (such as satisfaction scores, changes in health status, etc.) can be collected. This feedback information can then be used to further optimize the parameters (i.e., weighting coefficients) in the reward function via RLHF. For example, when a health intervention receives positive feedback, its weight can be increased to emphasize this aspect; conversely, if a health intervention is ineffective or user satisfaction is low, its weight can be decreased. This allows for continuous optimization of health intervention decisions, ensuring that interventions are both tailored to individual differences and follow physician guidance. This feedback loop makes health interventions more flexible, scientific, and effective, contributing to a better user experience in health management.

[0086] After step S50 in some embodiments, feedback information after the user takes health intervention measures can also be obtained; the medical big data model can be optimized based on the feedback information.

[0087] To ensure the long-term effectiveness of personalized interventions and optimize health intervention measures, online learning mechanisms can be used to optimize large-scale medical models.

[0088] The online learning mechanism continuously gathers feedback to dynamically update the parameters of the large-scale medical model, thereby optimizing personalized health interventions. The parameter update formula for the large-scale medical model is as follows:

[0089] in, The learning rate determines the magnitude of parameter adjustments to the large medical model during each update; These are the parameters of the current medical big data model; Based on Gradient updates represent updating the parameters of the large medical model in the direction of maximizing the reward.

[0090] Specifically, rewards can be calculated based on user feedback after each intervention (such as the degree of improvement in health indicators), and parameters can be adjusted through gradient updates. This parameter update method allows the large-scale medical model to continuously optimize as the user's health status changes, and it has long-term personalized adaptability.

[0091] When user feedback clearly indicates significant improvement in certain health indicators after intervention, the medical big data model will update its parameters in a direction that "generates better feedback." This ensures that the medical big data model learns and evolves, becoming increasingly adaptable to users. This allows the medical big data model to predict health risk probabilities more accurately, thereby generating more personalized health interventions based on these more accurate health risk probabilities, and further improving the effectiveness of health management.

[0092] Conversely, if it is found that certain health interventions are consistently ineffective or that a user's health condition does not improve significantly, they can be reported to the doctor for review. The doctor or other personnel will then adjust the parameters of the medical big data model based on the feedback and clinical experience to ensure that the health interventions are more precise and effective.

[0093] In this way, the parameters of the medical big data model are dynamically updated through real-time user feedback, enabling the model to continuously optimize as users' health status changes, thus ensuring long-term, adaptive health management support for users.

[0094] As can be seen, the above solution realizes an end-to-end real-time adaptive closed-loop health management mechanism, including: sensor acquisition - multimodal health feature fusion - health knowledge graph construction - health risk prediction - personalized intervention - user feedback - parameter update. This closed-loop mechanism means that the embodiments of the present invention not only passively monitor and remind users, but also proactively understand the user's health risk status, infer appropriate health intervention measures, and continuously optimize and evolve based on this. This ensures that health management for users not only provides personalized health intervention measures, but also adapts to user needs through continuous evolution, thereby achieving long-term effective health management and promoting the transformation from a "detection-reminder" to a "understanding-reasoning-intervention-evolution" health management mechanism.

[0095] The AI-based health management method provided in the above embodiments has the following effective effects: On the one hand, by acquiring multimodal fusion health features that comprehensively reflect the user's overall health status, a powerful, flexible health knowledge graph covering the causal relationships between relevant health events under the user's overall health status can be constructed, providing theoretical support and data basis for health management. On the other hand, knowledge subgraphs related to the multimodal fusion health features are retrieved from the health knowledge graph, enabling efficient and intelligent prediction of the user's health risks through a large medical model based on the multimodal fusion health features and knowledge subgraphs, thus obtaining the user's health risk probability. This allows the large medical model to comprehensively consider the overall health status reflected by the multimodal fusion health features and the key medical knowledge provided by the knowledge subgraphs when making risk predictions, improving the accuracy of health risk probabilities and enhancing the predictability of health management. Furthermore, by abstracting the user's health intervention into a reinforcement learning task, and using reinforcement learning algorithms, personalized health intervention measures are generated for the user based on the multimodal fusion health features, knowledge subgraphs, and health risk probabilities. This will provide accurate and scientific guidance for the user's health management, enhance the proactive intervention capability of health management, and significantly improve the accuracy, effectiveness, efficiency, and impact of health management.

[0096] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0097] It should be noted that the software tools or components not belonging to our company that appear in the embodiments of this invention are merely illustrative examples and do not represent actual use.

[0098] In one embodiment, an AI-based health management device is provided, which corresponds one-to-one with the AI-based health management method described in the above embodiments. As shown in Figure 4, the AI-based health management device includes a feature acquisition module 101, a graph construction module 102, a subgraph retrieval module 103, a risk prediction module 104, and a measure generation module 105. The functional modules are described in detail below: The feature acquisition module 101 is used to acquire the user's multimodal fusion health features; the graph construction module 102 is used to construct the user's health knowledge graph based on the multimodal fusion health features; the subgraph retrieval module 103 is used to retrieve knowledge subgraphs related to the multimodal fusion health features from the health knowledge graph; the risk prediction module 104 is used to predict the user's health risk based on the multimodal fusion health features and the knowledge subgraphs using a preset medical big data model, obtaining the user's health risk probability; the measure generation module 105 is used to generate health intervention measures for the user based on the multimodal fusion health features, the knowledge subgraphs, and the health risk probability using a reinforcement learning algorithm.

[0099] In one embodiment, the graph construction module 102 is specifically used for: constructing a health knowledge graph structure for the user; updating the health knowledge graph structure based on the multimodal fusion health features to obtain the health knowledge graph.

[0100] In one embodiment, the graph construction module 102 is further configured to: calculate a health risk score function based on the multimodal fusion health features; update the node weights of the health knowledge graph structure based on the health risk score function; and update the edge weights of the health knowledge graph structure based on a preset medical knowledge base, thereby obtaining the health knowledge graph.

[0101] In one embodiment, the risk prediction module 104 is specifically used to: concatenate the multimodal fusion features and the knowledge subgraph to obtain the user's health status representation; input the health status representation into the medical big data model for health risk prediction processing to obtain the health risk probability.

[0102] In one embodiment, the measure generation module 105 is specifically used to: integrate the multimodal fused health features, the knowledge subgraph, and the health risk probability into the user's health risk status; calculate the rewards of the candidate intervention measures available under the health risk status to obtain the rewards of the candidate intervention measures; and select the candidate intervention measure with the highest reward as the health intervention measure.

[0103] In one embodiment, the AI-based health management device further includes a model optimization module, which is used to: obtain feedback information after the user takes the health intervention measures; and optimize the medical big data model based on the feedback information.

[0104] In one embodiment, the feature acquisition module 101 is further configured to: acquire the user's multimodal health features; and perform fusion processing on the multimodal health features using a preset Transformer model to obtain the multimodal fused health features.

[0105] This invention provides an artificial intelligence-based health management device. On one hand, it acquires multimodal fusion health features that comprehensively reflect a user's overall health status to construct a powerful, flexible health knowledge graph encompassing causal relationships among relevant health events under the user's overall health status, providing theoretical support and data basis for health management. On the other hand, it retrieves knowledge subgraphs related to the multimodal fusion health features from the health knowledge graph. This allows a large-scale medical model to efficiently and intelligently predict the user's health risks based on the multimodal fusion health features and the knowledge subgraphs, obtaining the user's health risk probability. This enables the large-scale medical model to comprehensively consider the overall health status reflected by the multimodal fusion health features and the key medical knowledge provided by the knowledge subgraphs when making risk predictions, improving the accuracy of health risk probabilities and enhancing the predictability of health management. Furthermore, it abstracts health interventions for users into reinforcement learning tasks. Using reinforcement learning algorithms, it generates personalized health intervention measures for users based on multimodal fusion health features, knowledge subgraphs, and health risk probabilities. This provides precise and scientific guidance for user health management, enhances the proactive intervention capability of health management, and significantly improves the accuracy, effectiveness, efficiency, and impact of health management.

[0106] For specific limitations regarding AI-based health management devices, please refer to the limitations of AI-based health management methods described above, which will not be repeated here. The modules in the aforementioned AI-based health management device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0107] In one embodiment, a computer device is provided, which can be a server, and its internal structure diagram is shown in Figure 5. The computer device includes a processor, memory, a network interface, and a database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side health management method based on artificial intelligence.

[0108] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: acquiring a user's multimodal fusion health features; constructing a health knowledge graph of the user based on the multimodal fusion health features; retrieving knowledge subgraphs related to the multimodal fusion health features from the health knowledge graph; predicting the user's health risk based on the multimodal fusion health features and the knowledge subgraphs using a preset medical big data model to obtain the user's health risk probability; and generating health intervention measures for the user based on the multimodal fusion health features, the knowledge subgraphs, and the health risk probability using a reinforcement learning algorithm.

[0109] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program performs the following steps: acquiring a user's multimodal fusion health features; constructing a health knowledge graph of the user based on the multimodal fusion health features; retrieving knowledge subgraphs related to the multimodal fusion health features from the health knowledge graph; predicting the user's health risk based on the multimodal fusion health features and the knowledge subgraphs using a preset medical big data model to obtain the user's health risk probability; and generating health intervention measures for the user based on the multimodal fusion health features, the knowledge subgraphs, and the health risk probability using a reinforcement learning algorithm.

[0110] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0112] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0113] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A health management method based on artificial intelligence, characterized in that, include: Obtain the user's multimodal fusion health characteristics; Construct the user's health knowledge graph based on the multimodal fusion health features; Retrieve knowledge subgraphs related to the multimodal fusion health features from the health knowledge graph; use a preset medical big model to predict the user's health risk based on the multimodal fusion health features and the knowledge subgraphs to obtain the user's health risk probability; Using reinforcement learning algorithms, health intervention measures are generated for the user based on the multimodal fusion health features, the knowledge subgraph, and the health risk probability.

2. The health management method as described in claim 1, characterized in that, The step of constructing the user's health knowledge graph based on the multimodal fusion health features includes: constructing a health knowledge graph structure for the user; and updating the health knowledge graph structure based on the multimodal fusion health features to obtain the health knowledge graph.

3. The health management method as described in claim 2, characterized in that, The step of updating the health knowledge graph structure based on the multimodal fusion health features to obtain the health knowledge graph includes: calculating a health risk score function based on the multimodal fusion health features; updating the node weights of the health knowledge graph structure based on the health risk score function; and updating the edge weights of the health knowledge graph structure based on a preset medical knowledge base to obtain the health knowledge graph.

4. The health management method as described in claim 1, characterized in that, The step of using a pre-defined medical big data model to predict the user's health risk based on the multimodal fusion health features and the knowledge subgraph to obtain the user's health risk probability includes: splicing the multimodal fusion features and the knowledge subgraph to obtain the user's health status representation; and inputting the health status representation into the medical big data model for health risk prediction to obtain the health risk probability.

5. The health management method as described in claim 1, characterized in that, The step of generating health intervention measures for the user based on the multimodal fusion health features, the knowledge subgraph, and the health risk probability using a preset reinforcement learning algorithm includes: integrating the multimodal fusion health features, the knowledge subgraph, and the health risk probability into the user's health risk status; calculating the rewards of the candidate intervention measures available under the health risk status to obtain the rewards of the candidate intervention measures; and selecting the candidate intervention measure with the highest reward as the health intervention measure.

6. The health management method according to any one of claims 1 to 5, characterized in that, After generating health intervention measures for the user based on the multimodal fusion health features, the knowledge subgraph, and the health risk probability using a reinforcement learning algorithm, the method further includes: obtaining feedback information after the user takes the health intervention measures; and optimizing the medical big data model based on the feedback information.

7. The health management method according to any one of claims 1 to 5, characterized in that, The step of obtaining the user's multimodal fused health features includes: obtaining the user's multimodal health features; and fusing the multimodal health features using a preset Transformer model to obtain the multimodal fused health features.

8. A health management device based on artificial intelligence, characterized in that, include: The feature acquisition module is used to acquire the user's multimodal fusion health features; The knowledge graph construction module is used to construct the user's health knowledge graph based on the multimodal fused health features; The subgraph retrieval module is used to retrieve knowledge subgraphs related to the multimodal fusion health features from the health knowledge graph; the risk prediction module is used to predict the user's health risk based on the multimodal fusion health features and the knowledge subgraphs using a preset medical big model, and obtain the user's health risk probability. The measure generation module is used to generate health intervention measures for the user based on the multimodal fusion health features, the knowledge subgraph, and the health risk probability using a reinforcement learning algorithm.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the artificial intelligence-based health management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the artificial intelligence-based health management method as described in any one of claims 1 to 7.

Images