Health monitoring system and method based on multi-source knowledge data fusion

Abstract

The invention discloses a health monitoring system and method based on multi-source knowledge data fusion, and the system comprises a server side and an edge computing node, the edge computing node is used for interacting with a user, collecting user data, feeding back a result, and transmitting the user data to the server side after anonymization; and the server is used for calculating according to the anonymous user data to obtain a health monitoring result and returning the health monitoring result to the edge computing node. The method is realized by adopting the system. According to the method, the accuracy of personal health condition monitoring evaluation and disease risk prediction is improved through multi-source knowledge data fusion, and the personal health data development and utilization under the condition that both the personal privacy data and the service provider algorithm model are not leaked are realized through bidirectional shielding of the personal health data and the health monitoring model. Through reasonable model splitting, the operand of the server and the edge computing node can be efficiently distributed, and the overall operation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to a health monitoring system and method, in particular to a health monitoring system and method based on multi-source knowledge data fusion. Background Art

[0002] This section merely provides background information related to the present disclosure and is not necessarily prior art.

[0003] With the improvement of quality of life and the development of technology, the traditional medical service model cannot meet people's health needs. Personal health management based on big data technology has become increasingly important, that is, providing personalized health services by collecting and analyzing large amounts of personal health data, such as early detection and prevention of diseases, real-time tracking of health status, and providing health protection advice and medical decision support.

[0004] However, this process often involves the collection, use and processing of personal sensitive information, such as genetic information, disease information, biometric data, etc. The conflict between these behaviors and personal privacy protection is becoming increasingly prominent. Although there are currently relevant laws, regulations and technical means to protect personal privacy data to a certain extent, it is still difficult to resolve the contradiction between data utilization and privacy protection. Existing cases have shown that even if privacy protection measures such as anonymization and data encryption are taken, it is still difficult to completely avoid the risk of privacy leakage.

[0005] On the other hand, from the perspective of data service providers, algorithm models and their parameters are often important trade secrets of the company, especially those models that can provide unique value and services after being trained with a large amount of data, which usually require a lot of capital, manpower and time costs. While companies use these algorithm models and their parameters to provide external services, protecting these trade secrets is crucial to maintaining the company's competitive advantage.

[0006] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0007] Purpose of the invention: The technical problem to be solved by the present invention is to provide a health monitoring system and method based on multi-source knowledge data fusion in view of the deficiencies in the prior art.

[0008] In order to solve the above technical problems, the present invention discloses a health monitoring system and method based on multi-source knowledge data fusion, wherein the system comprises:

[0009] The server and the edge computing node, wherein the edge computing node is used to interact with the user, collect user data and provide result feedback, and send the user data to the server after anonymization; the server is used to perform calculations based on the anonymized user data, obtain health monitoring results and return them to the edge computing node.

[0010] Furthermore, the server includes a model training module and a model splitting module; wherein the model training module is used to train and obtain a health monitoring model, and the model splitting module splits the health monitoring model to obtain an inference calculation model and a knowledge extraction model, which are respectively deployed in the server and the edge computing node to jointly complete the health monitoring of the user.

[0011] Furthermore, the edge computing node includes a data access module and a data anonymization module; wherein the data access module is used to collect user data, and the data anonymization module is used to anonymize the data.

[0012] Furthermore, the inference computing model is deployed on the server side and is used to perform inference computing based on the data sent by the edge computing node.

[0013] Furthermore, the knowledge extraction model is deployed on an edge computing node, and the knowledge data used for the inference computing model to perform inference computing is extracted from the user data collected by the data access module.

[0014] Furthermore, the knowledge extraction model and the reasoning calculation model interact through a data anonymization module.

[0015] Furthermore, the interaction through the data anonymization module specifically includes:

[0016] The knowledge extraction model sends the knowledge data to the data anonymization module, anonymizes the knowledge data, and then sends it to the reasoning calculation model;

[0017] The inference calculation model performs inference calculation based on the acquired anonymized knowledge data to obtain anonymous health monitoring results, and sends them to the data anonymization module, which converts them into health monitoring results for users.

[0018] The present invention also proposes a health monitoring method based on multi-source knowledge data fusion, which uses the above system to perform health monitoring, including the following steps:

[0019] Step 1: Obtain a health monitoring model, wherein the health monitoring model performs health status assessment and risk prediction based on input data;

[0020] Step 2, splitting the health monitoring model into a reasoning calculation model and a knowledge extraction model;

[0021] Step 3: deploy the reasoning computing model and the knowledge extraction model in the server and edge computing node respectively;

[0022] Step 4: Collect user data, extract knowledge, and obtain knowledge data;

[0023] Step 5, anonymizing the knowledge data and sending it to the inference calculation model for processing;

[0024] Step 6, the inference calculation model is used to process and obtain anonymous health monitoring results;

[0025] Step 7: de-anonymize the anonymous health monitoring results to obtain the user's final health monitoring results.

[0026] Furthermore, the step of obtaining the health monitoring model in step 1 includes:

[0027] Step 1-1, using supervised learning or deep learning methods to build a deep artificial neural network;

[0028] Step 1-2, collecting a health monitoring training data set and training the deep artificial neural network;

[0029] Step 1-3, using the trained deep artificial neural network as a health monitoring model.

[0030] Furthermore, the health monitoring training data set described in step 1 includes:

[0031] Medical data, physiological indicator data, behavioral data, psychological indicator data, genetic factor data and environmental factor data.

[0032] Beneficial effects:

[0033] 1. Collect and integrate multiple types of personal health data from multiple data sources for cross-domain integration and correlation analysis to provide more comprehensive health monitoring and help discover health problems that are difficult to capture with a single data source.

[0034] 2. The original data is not submitted to the server. Instead, the knowledge data is extracted locally and anonymized before being submitted to the server to avoid the risk of user data leakage;

[0035] 3. The calculation model and weight parameters do not need to be deployed to the user end to protect the business secrets of the service provider.

[0036] 4. Through reasonable model splitting, the computing workload of the server and edge computing nodes is reasonably allocated to improve the overall computing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0038] Figure 1 It is a schematic diagram of the system structure of the present invention.

[0039] Figure 2 Schematic diagram of the health monitoring model and its split deployment. DETAILED DESCRIPTION

[0040] The present invention proposes a method and system for health monitoring based on multi-source knowledge data fusion. With the help of this method and system, personal health data from multiple data sources can be collected for health status monitoring, while protecting the privacy of users and the commercial secrets of service providers such as algorithm models and parameters from being leaked.

[0041] In order to achieve the above object, the technical solution of the present invention is achieved as follows:

[0042] A health monitoring method and system based on multi-source knowledge data fusion, comprising:

[0043] The server is an application managed by the service provider and runs on the cloud or central server. It is used for: 1. In the model training phase, the collected personal health data set is used to train a health monitoring model that can assess health status, predict disease risks, and give behavioral recommendations; 2. In the model deployment phase, the split learning technology is used to split the model and deploy it locally on the server and on the edge computing node; 3. In the inference phase, according to the health monitoring model, the edge computing node and the server-side model are coordinated to jointly assess personal health risks and give behavioral recommendations. It includes a health monitoring model, a model splitting module, and an inference computing model.

[0044] Edge computing nodes are applications running in the user’s local environment or device, through which sensitive data can be processed locally without transmitting the data to the cloud or central server, thus avoiding the risk of data leakage during transmission or after transmission to the service provider.

[0045] The health monitoring model is a set of calculation functions and their weight parameters obtained by training based on medical diagnosis and scientific research data through machine learning algorithm models such as neural networks, analyzing the relationship between multiple influencing factors and diseases or health risks. By inputting personal health data into this model, it is possible to monitor personal health status, provide health risk assessment results, and give behavioral recommendations in terms of medical treatment, diet, exercise, sleep, etc.

[0046] The personal health data include: medical data such as diagnosis and medication; physiological indicator data such as body temperature, heart rate, blood pressure, respiratory rate, blood sugar level, etc.; behavioral data such as sleep, diet, exercise and other behavioral records and behavioral characteristics; psychological indicator data such as mental illness, stress or emotional state; genetic factor data such as genetic diseases, family medical history, individual susceptibility to diseases caused by genes carried, etc.; environmental factor data such as natural factors such as temperature, humidity, air quality, noise, radiation, as well as chemical factors such as polluted water quality, toxic and harmful gases, or biological factors such as various pathogenic microorganisms.

[0047] The model splitting module splits the model according to the characteristics of the machine learning algorithm. For example, the neural network model algorithm is calculated in layers, that is, the model includes an input layer, several hidden layers and an output layer, and each layer contains multiple neurons. For the trained model, the model can be divided into multiple parts according to horizontal splitting, vertical splitting or mixed splitting methods, and different parts are assigned to different devices for operation.

[0048] For the health monitoring model, a vertical splitting method can be used to split the health monitoring model f(x, θ) trained in advance by the service provider into a knowledge extraction model f1 and an inference calculation model f2. The knowledge extraction model f1 is sent to the edge computing node, and the inference calculation model f2 is processed on the server side.

[0049] If the personal health data collected by the edge computing node is x, the output of the edge computing node is:

[0050] h=f1(x,θ1)

[0051] where θ1 is the model parameter of f1.

[0052] The calculation results of the edge computing node are returned to the server as the input of the inference computing model, and the calculation results are:

[0053] y=f2(h,θ2)

[0054] Among them, θ2 is the model parameter of f2.

[0055] The edge computing node includes:

[0056] The data access module is used to define the data interface with the third party and receive personal health monitoring-related data sent by different data sources such as third-party applications and smart devices according to user authorization;

[0057] The knowledge extraction model is used to calculate the data related to personal health monitoring, extract the knowledge data that does not directly contain personal privacy information, and send it to the inference calculation model;

[0058] The data anonymization module is used to realize anonymous communication between edge computing nodes and the server, ensuring that the knowledge data sent to the server does not contain node identity information.

[0059] The data access module comprises:

[0060] Data interface, used to receive personal health data submitted by various data sources, and to enable data sources to connect and submit data by agreeing on communication protocols, data formats, encoding methods, security guarantees, etc.;

[0061] Metadata management is used to uniformly manage the model definition and storage of metadata related to personal health. The metadata is used to identify and manage the accessed data, and the identified data is used as the input of the corresponding calculation model.

[0062] Data preprocessing involves cleaning and converting the received personal health data to ensure the accuracy and consistency of the data.

[0063] The data source mentioned here refers to the software or hardware that collects, stores or manages the user's personal health data.

[0064] The data anonymization module can use stateless session management and encryption technology to achieve the goal of not providing any edge computing node identity and status information to the server, and still using the services provided by the server. And through the content security policy, it is ensured that the content returned by the server is safe and will not be obtained by a third party.

[0065] Embodiment 1:

[0066] Figure 1 The figure shows a schematic diagram of the structure of a health monitoring system based on multi-source knowledge data fusion implemented according to the present invention, which specifically includes:

[0067] On the server side, an application running on the cloud or central server managed by the service provider includes a health monitoring model, a model splitting module, and an inference computing model.

[0068] Edge computing nodes are applications running in the user’s local environment or device, through which sensitive data can be processed locally without transmitting the data to the cloud or central server, thus avoiding the risk of data leakage during transmission or after transmission to the service provider.

[0069] In this embodiment, during the model training phase, after collecting personal health data, a deep artificial neural network can be constructed using supervised learning or deep learning algorithms, and a set of calculation functions and their weight parameters obtained by training and analyzing the correlation between multiple influencing factors and health risks, namely the health monitoring model, can be obtained. Based on this model, personal health status can be monitored, and the impact of different factors on health status can be analyzed, so as to provide health monitoring results such as disease risk prediction and auxiliary disease diagnosis, as well as behavioral suggestions that are beneficial to physical health in terms of medical treatment, diet, exercise, sleep, etc.

[0070] The personal health data in this embodiment includes: medical data such as diagnosis and medication data; physiological indicators such as body temperature, heart rate, blood pressure, respiratory rate, blood sugar level and other data; behavioral data such as sleep, diet, exercise and other behavioral records and behavioral characteristics; psychological indicators such as mental illness or emotional state data; genetic factors such as genetic diseases, family medical history, and disease susceptibility caused by individual genes; environmental factors such as temperature, humidity, air quality, noise, radiation and other natural factors; as well as chemical factors such as polluted water quality, toxic and harmful gases, or biological factors such as various pathogenic microorganisms.

[0071] In this embodiment, during the model deployment phase, the model splitting module is used to convert the health monitoring model and weight parameters pre-trained by the service provider to generate a knowledge extraction model and an inference calculation model, and the knowledge extraction model is sent to the edge computing node respectively; the knowledge extraction model is used to process the user's personal health data locally to obtain knowledge data and return it to the server, and the inference calculation model of the server takes the knowledge data as input to obtain the user's health status monitoring results, and gives disease risk assessment and behavior recommendations. In addition, the splitting process does not affect the health status monitoring results, and the results are the same as those obtained by taking personal health data as the input of the health monitoring model.

[0072] The conversion process is irreversible, that is, the original health monitoring model cannot be inferred based on the knowledge extraction model or the reasoning calculation model, so as to protect the health monitoring model from being copied or reverse engineered.

[0073] In this embodiment, the deep learning neural network can be horizontally split, and the knowledge extraction model after the split can be deployed on the edge computing node for calculation. At the same time, privacy computing technology is used to solve the trust problem of edge computing nodes in the server. The edge computing nodes in this embodiment can be deployed in users' mobile phones, computers and other devices, including: a data access module, which is used to define the data interface with a third party, and receive personal health monitoring related data sent by different data sources such as third-party applications and smart devices according to user authorization; a knowledge extraction model, which is used to calculate the data related to personal health monitoring, extract knowledge data that does not directly contain personal privacy information, and send it to the reasoning calculation model; a data anonymization module, which is used to realize anonymous communication between the edge computing node and the server, and ensure that the knowledge data sent to the server does not contain node identity information.

[0074] The data access module includes: a data interface for receiving personal health data submitted by various data sources, and providing data sources with docking and data submission through agreed communication protocols, data formats, encoding methods, security guarantees, etc.; metadata management for unified management of model definitions and storage of metadata related to personal health, and identification and management of accessed data through metadata, and using the identified data as input for the corresponding calculation model. Data preprocessing: cleaning and conversion of received personal health data to ensure data accuracy and consistency.

[0075] The source of personal health data refers to the software or hardware that collects, stores or manages the user's personal health data and its influencing factors. It can be a file or software system that records or stores personal health data, as well as wearable devices, medical devices, or environmental sensors that can collect human body data, social media that reflects personal stress and psychological state, etc.

[0076] In this embodiment, the data access module is used to connect to known applications or devices containing the user's personal health data to collect the user's personal health data. After the data collection is completed, the definitions and constraints of various data elements are checked and preprocessed according to the metadata management module. First, noise and outliers are removed through data cleaning to ensure data quality, and then the data is standardized to convert data from different sources into a unified format and unit. Finally, multi-source data is fused to integrate and unify data from different data sources.

[0077] The knowledge extraction model includes a knowledge extraction step and a feature extraction step. In the knowledge extraction stage, useful knowledge entities such as average heart rate, exercise steps, sleep quality index, etc. are extracted from the original data. This step can effectively reduce the number of model inputs. In the feature extraction step, a set of feature calculation functions sent by the server is used, and the knowledge entity is used as the input of the set of calculation functions to generate the corresponding intermediate feature representation of the knowledge data. And the conversion process is irreversible, that is, it is impossible to infer a unique input data combination based on the output value, thereby ensuring that the output data does not leak any personal privacy information.

[0078] The data anonymization module of this embodiment can use stateless session management and encryption technology to achieve the goal of not providing the server with the identity and status information of any edge computing node, and still using the services provided by the server. And through the content security policy, it is ensured that the content returned by the server is safe and will not be obtained by a third party. In this embodiment, multi-party secure computing or homomorphic encryption technology can be used to achieve this anonymization effect.

[0079] In this embodiment, after the server receives and decrypts the knowledge data sent by the edge computing node, it uses the deployed inference computing model to further process the user's health knowledge data to generate the final prediction result. The prediction result is then encrypted and returned to the client, which decrypts the prediction result and displays it to the user. In this process, the data sent by the edge computing node to the server does not contain sensitive information. The server can obtain health status monitoring results based on the knowledge data, but cannot associate the results with any user or device; the service provider does not need to deploy the model to the user environment, which avoids the model from being copied or reverse engineered. This achieves two-way shielding of data and models between the edge computing node and the server.

[0080] The health monitoring method and system based on multi-source knowledge data fusion described in the present invention collects and integrates multiple types of personal health data from multiple data sources for cross-domain integration and correlation analysis, provides more comprehensive health monitoring, and helps to discover health problems that are difficult to capture with a single data source. Neither the individual users of the system nor the health monitoring service providers obtain each other's original data, but use the converted knowledge data and the converted model to calculate and obtain personal health monitoring results. The data and model are bidirectionally shielded to achieve the development and utilization of personal health data without leaking personal privacy data and commercial secrets such as the algorithm model of the service provider. Through reasonable model splitting, the computing power of the server and edge computing nodes can be efficiently allocated, improving the overall computing efficiency.

[0081] Embodiment 2:

[0082] Figure 2The present invention is based on the health monitoring model based on multi-source knowledge data fusion and its split deployment schematic diagram, the health monitoring model includes an input layer, an output layer and n hidden layers. Among them, the input layer is used to receive personal health data, and the output layer is a classification label related to several diseases or health risks. After training, the model can reflect the correlation between personal health status and health data, and then evaluate the health status or predict disease risks based on personal health data, and further give behavioral suggestions that are beneficial to physical health.

[0083] In this embodiment, the health monitoring model is split by layer to obtain a knowledge extraction model f1 and an inference calculation model f2, and the knowledge extraction model f1 can be further split into f according to the type of data contained in each edge computing node. 11 ……f 1i . The knowledge extraction model after splitting f 1i Deployed on the i-th edge computing node, f2 is deployed on the server.

[0084] For the i-th edge computing node, the personal health data collected or aggregated is used as input, and the calculation results are transmitted to the server to complete the calculation of the remaining network, and then returned to the edge computing node to calculate the prediction results, and finally complete a prediction process. Alternatively, the server initiates calculation instructions to each edge computing node, collects the knowledge data extracted by each edge computing node, and then performs reasoning to obtain the final evaluation result.

[0085] In a specific implementation, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, the invention content of a health monitoring system and method based on multi-source knowledge data fusion provided by the present invention and some or all steps in each embodiment can be executed. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0086] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of computer programs and their corresponding general hardware platforms. Based on such an understanding, the technical solutions in the embodiments of the present invention can be essentially or partly contributed to the prior art in the form of computer programs, i.e., software products, which can be stored in a storage medium and include several instructions for enabling a device including a data processing unit (which can be a personal computer, a server, a single-chip microcomputer, an MCU or a network device, etc.) to execute the methods described in various embodiments of the present invention or certain parts of the embodiments.

[0087] The present invention provides a health monitoring system and method based on multi-source knowledge data fusion. There are many methods and ways to implement the technical solution. The above is only a preferred implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the scope of protection of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A health monitoring system based on multi-source knowledge data fusion, characterized in that: include: The server and the edge computing node, wherein the edge computing node is used to interact with the user, collect user data and provide result feedback, and send the user data to the server after anonymization; the server is used to perform calculations based on the anonymized user data, obtain health monitoring results and return them to the edge computing node.

2. A health monitoring system based on multi-source knowledge data fusion according to claim 1, characterized in that: The server includes a model training module and a model splitting module; wherein the model training module is used to train and obtain a health monitoring model, and the model splitting module splits the health monitoring model to obtain an inference calculation model and a knowledge extraction model, which are respectively deployed in the server and the edge computing node to jointly complete the health monitoring of the user.

3. A health monitoring system based on multi-source knowledge data fusion according to claim 2, characterized in that: The edge computing node includes a data access module and a data anonymization module; wherein the data access module is used to collect user data, and the data anonymization module is used to anonymize the data.

4. A health monitoring system based on multi-source knowledge data fusion according to claim 3, characterized in that: The inference computing model is deployed on the server side and is used to perform inference computing based on the data sent by the edge computing node.

5. A health monitoring system based on multi-source knowledge data fusion according to claim 4, characterized in that: The knowledge extraction model is deployed on the edge computing node, and extracts knowledge data used for the inference computing model to perform inference computing from the user data collected by the data access module.

6. A health monitoring system based on multi-source knowledge data fusion according to claim 5, characterized in that: The knowledge extraction model and the reasoning calculation model interact through a data anonymization module.

7. A health monitoring system based on multi-source knowledge data fusion according to claim 6, characterized in that: The interaction through the data anonymization module specifically includes: The knowledge extraction model sends the knowledge data to the data anonymization module, anonymizes the knowledge data, and then sends it to the reasoning calculation model; The inference calculation model performs inference calculation based on the acquired anonymized knowledge data to obtain anonymous health monitoring results, and sends them to the data anonymization module, which converts them into health monitoring results for users.

8. A health monitoring method based on multi-source knowledge data fusion, characterized in that: Using the system described in claims 1-7 to perform health monitoring comprises the following steps: Step 1: Obtain a health monitoring model, wherein the health monitoring model performs health status assessment and risk prediction based on input data; Step 2, splitting the health monitoring model into a reasoning calculation model and a knowledge extraction model; Step 3: deploy the reasoning computing model and the knowledge extraction model in the server and edge computing node respectively; Step 4: Collect user data, extract knowledge, and obtain knowledge data; Step 5, anonymizing the knowledge data and sending it to the inference calculation model for processing; Step 6, the inference calculation model is used to process and obtain anonymous health monitoring results; Step 7: de-anonymize the anonymous health monitoring results to obtain the user's final health monitoring results.

9. The health monitoring method based on multi-source knowledge data fusion according to claim 8, characterized in that: The step 1 of obtaining the health monitoring model includes: Step 1-1, using supervised learning or deep learning methods to build a deep artificial neural network; Step 1-2, collecting a health monitoring training data set and training the deep artificial neural network; Step 1-3, using the trained deep artificial neural network as a health monitoring model.

10. The health monitoring method based on multi-source knowledge data fusion according to claim 9, characterized in that: The health monitoring training dataset described in step 1 includes: Medical data, physiological indicator data, behavioral data, psychological indicator data, genetic factor data and environmental factor data.

Images