Intelligent health care system

Abstract

The invention belongs to the technical field of health service, and discloses an intelligent health care system, which comprises a client used for collecting physiological feature data and posture feature data, predicting corresponding symptoms by using a prediction model based on the physiological feature data and the posture feature data, and matching health care equipment parameters corresponding to the symptoms; the prediction model is constructed based on a neural network; the equipment control unit is used for controlling the corresponding health care equipment according to the health care equipment parameters; and the distributed optimization unit is used for learning the health care features of the newly registered users from a plurality of clients in an aggregation mode, and updating the prediction model of each client. According to the technical scheme, autonomous decision making and continuous evolution of the health care equipment can be achieved, and meanwhile user privacy safety is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of health services, and particularly relates to an intelligent health and rehabilitation system. BACKGROUND

[0002] Current intelligent health and rehabilitation systems mainly collect single physiological data through wearable devices such as wristbands and chest bands, combine simple threshold rules or traditional machine learning models to conduct health early warning, and control health and rehabilitation devices through preset programs. Some systems attempt to introduce multi-modal perception, but it is difficult to implement due to privacy issues, and device parameter adjustment relies on human experience, lacking dynamic adaptability. SUMMARY

[0003] The purpose of the present application is to provide an intelligent health and rehabilitation system to solve the problems existing in the prior art.

[0004] To achieve the above purpose, the present application provides an intelligent health and rehabilitation system, comprising: a client for collecting physiological feature data and posture feature data, predicting corresponding symptoms based on the physiological feature data and the posture feature data using a prediction model, and matching health and rehabilitation device parameters corresponding to the symptoms; the prediction model is constructed based on a neural network; a device control unit for controlling corresponding health and rehabilitation devices according to health and rehabilitation device parameters; a distributed optimization unit for learning health and rehabilitation features of new registered users from multiple clients in an aggregated manner, and updating the prediction model of each client.

[0005] Optionally, the client specifically comprises: a feature monitoring and collecting module for collecting physiological feature data and posture feature data; a data analysis module for predicting corresponding symptoms using a prediction model based on the physiological feature data and the posture feature data, and matching health and rehabilitation device parameters corresponding to the symptoms in combination with a knowledge graph.

[0006] Optionally, the feature monitoring and collecting module specifically comprises: a physiological feature data collecting submodule for collecting heart rate, blood pressure and blood oxygen data of a user using a flexible sensor to obtain physiological feature data; a posture feature data collecting submodule for acquiring posture feature data of a user using a millimeter wave radar.

[0007] Optionally, the feature monitoring and collecting module further comprises: a data preprocessing module for preprocessing collected data based on a sensor space-time alignment algorithm.

[0008] Optionally, the data analysis module specifically comprises: The model construction submodule is configured to construct a prediction model, wherein the prediction model comprises a feature extraction layer and a personalized prediction layer connected in sequence, the backbone network of the feature extraction module is a ResNet-18 network, the auxiliary network is a BiLSTM network, and the personalized prediction layer is composed of a full connection layer and an activation function. The symptom prediction submodule is configured to input the physiological feature data and the posture feature data into the prediction model for classification prediction to obtain a symptom prediction result. The knowledge graph matching submodule is configured to generate device control parameters according to the symptom prediction result by matching a knowledge graph.

[0009] Optionally, the distributed optimization unit specifically comprises: The initialization module is configured to initialize a federated learning training environment by a central server, generate an initial prediction model, and set a training round regulation value. The pre-training module is configured to pre-train each module in the initial prediction model by the central server, and distribute the pre-trained modules to each client. The client training module is configured to train each pre-trained module by the client to obtain an updated feature extraction module and an updated personalized module, train the updated personalized module by the client, obtain a separately trained personalized module, and save the separately trained personalized module in the client. The feature extraction module aggregation module is configured to send the updated feature extraction module to the central server for weighted average calculation, determine whether the current training round reaches the set training round regulation value by the central server, and determine whether the training is completed, if the regulation value is reached, the training is completed, the weighted average calculated feature extraction module is sent to the client, and the client is notified to stop training, and the client performs symptom prediction by using the separately trained personalized module and the weighted average calculated feature extraction module.

[0010] Optionally, the pre-training module specifically comprises: The dataset construction module is configured to create a medical dataset by using a generative adversarial network. The training module is configured to input the medical dataset into the initial prediction model for classification prediction, and train the initial prediction model according to a target loss function to obtain a pre-trained prediction model.

[0011] Optionally, the client training module specifically comprises: The data acquisition module is configured to acquire local data of the client. The training module is used for training each module after pre-training according to local data, obtaining an updated feature extraction module and an updated individualization module; the updated individualization module is separately trained based on small sample meta-learning, and the separately trained individualization module is obtained and saved in the client.

[0012] The technical effect of the present application is: The present application constructs a full-automatic closed-loop process of perception-decision-execution, significantly improves the symptom recognition accuracy by fusing millimeter wave radar attitude data and flexible sensor physiological data, and combining ResNet-18 and BiLSTM double network feature extraction. And the present application combines the federal learning architecture, the user original data is locally retained, and only the encrypted model parameters are uploaded, which meets the privacy requirements and realizes dynamic updating. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0014] The drawings forming part of the present application are used to provide further understanding of the present application, the illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute improper limitation on the present application. In the drawings: Figure 1 The system schematic diagram in the embodiments of the present application. DETAILED DESCRIPTION

[0015] Now, various exemplary embodiments of the present application will be described in detail, which should not be considered as limiting the present application, but should be understood as a more detailed description of some aspects, characteristics and embodiments of the present application.

[0016] It should be understood that the terms described in the present application are only for describing the specific embodiments, and are not used to limit the present application. In addition, for the numerical range in the present application, it should be understood that each intermediate value between the upper limit and the lower limit of the range is also specifically disclosed. Each smaller range between any stated value or intermediate value in the range and any other stated value or intermediate value in the range is also included in the present application. The upper limit and the lower limit of these smaller ranges can be independently included or excluded from the range.

[0017] Various modifications and variations can be made to the specific embodiments described in this specification without departing from the scope or spirit of the invention, as will be apparent to those skilled in the art. Other embodiments derived from this specification will also be obvious to those skilled in the art. This application specification and embodiments are merely exemplary.

[0018] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] like Figure 1 As shown, this embodiment provides an intelligent health and wellness system, including: a client, used to collect physiological characteristic data and posture characteristic data, and based on the physiological characteristic data and posture characteristic data, use a prediction model to predict corresponding symptoms and match the health and wellness equipment parameters corresponding to the symptoms; the prediction model is based on a neural network; an equipment control unit, used to control the corresponding health and wellness equipment according to the health and wellness equipment parameters; and a distributed optimization unit, used to learn the health and wellness characteristics of newly registered users from multiple clients in an aggregated manner and update the prediction model of each client.

[0021] This embodiment discloses an intelligent healthcare system based on federated learning and multimodal perception, including a client, a device control unit, and a distributed optimization unit. The client collects user physiological characteristics (heart rate / blood pressure / blood oxygen) and posture feature data through the fusion of flexible sensors and millimeter-wave radar, uses a ResNet-18+BiLSTM dual-network prediction model for symptom identification, and dynamically matches healthcare equipment parameters with a knowledge graph; the device control unit executes parameter commands in real time; the distributed optimization unit adopts a federated learning framework, aggregating the feature extraction layer parameters of multiple clients to optimize the global model, while retaining the local training of personalized prediction layers.

[0022] The specific implementation process of this embodiment includes: Data Acquisition: The user wears a flexible chest strap to collect heart rate / blood pressure / blood oxygen data at a sampling rate of 10Hz; a millimeter-wave radar is deployed on the ceiling to capture attitude point cloud data at 30fps. After data acquisition, preprocessing is performed using a sensor spatiotemporal alignment algorithm.

[0023] The preprocessed data is input into the prediction model for classification prediction. The prediction model consists of a feature extraction layer connected in sequence and a personalized prediction layer composed of a fully connected layer and an activation function. The classification prediction process is as follows: Feature extraction: ResNet-18 processes physiological data to obtain spatial features; BiLSTM bidirectional long short-term memory network analyzes posture time sequence features to output time sequence feature vectors.

[0024] The spatial feature vector and the time sequence feature vector are weighted and spliced through an attention mechanism to form a fusion feature.

[0025] The fusion feature is input into the personalized prediction layer, and the symptom probability is output based on the full connection layer; The corresponding symptoms are predicted using the prediction model, and the corresponding health and rehabilitation equipment parameters are matched by combining the knowledge graph; wherein the knowledge graph stores the symptom-equipment parameter mapping rule.

[0026] Perform federated learning optimization: The central server is initialized, 100,000 synthetic medical data (physiological + posture labels) are generated using GAN, the prediction model is pre-trained, and is distributed to each client; wherein the central server only sends the personalized prediction layer in the first round of training, and only sends the feature extraction layer in the remaining training.

[0027] Local training: the client trains the feature extraction layer and the personalized prediction layer based on local data and target loss function, obtains the updated feature extraction module and the updated personalized module; after the global model is updated, the updated personalized module is trained separately in the client using small sample meta-learning, and the separately trained personalized module is obtained and saved in the client; Aggregated update: the client uploads the updated feature extraction layer parameters to the server, and the weighted average is returned; the aggregation weight of the federated learning is upgraded to the data quality score in the system, and the data integrity, timeliness, and annotation quality are comprehensively considered to generate the score, and the central server dynamically adjusts the client weight according to the quality score, so as to filter noise data from the source and improve the accuracy of the model.

[0028] When the training round reaches the preset value, the client fixes the feature extraction layer and only retains the local personalized layer for prediction, i.e. using the separately trained personalized module and the weighted average calculated feature extraction module to predict symptoms.

[0029] After each round of aggregated update, the central server tests the global model using the validation set, and detects the model deviation in real time through cosine similarity, including: Parameter comparison: the local model parameters of the client are vectorized with the global model, and the cosine similarity is calculated; when the similarity is <0.85, the aggregation of the abnormal client is suspended, the learning rate is dynamically adjusted in the interval of 0.85-0.92, and millisecond-level fault localization can be achieved, To sum up, the embodiment constructs a full-automatic closed-loop process of perception-decision-execution, significantly improves the symptom recognition accuracy by fusing the millimeter wave radar attitude data and the flexible sensor physiological data, combining ResNet-18 and BiLSTM double network feature extraction, and combining the federated learning architecture, the original data of the user is locally retained, only the encrypted model parameters are uploaded, the privacy requirements are met while dynamic updating is realized.

[0030] The embodiment solves the problems of multi-modal data fusion analysis, new user cold start and privacy protection, significantly improves the symptom recognition accuracy and the device control individualization level, and is suitable for home and institutional health care scenes.

[0031] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent health and wellness system, characterized in that, The method comprises the steps of: a client, for collecting physiological feature data and posture feature data, predicting corresponding symptoms based on the physiological feature data and the posture feature data by using a prediction model, and matching health and rehabilitation equipment parameters corresponding to the symptoms; the prediction model is constructed based on a neural network; a device control unit, for controlling the corresponding health and rehabilitation equipment according to the health and rehabilitation equipment parameters; a distributed optimization unit, for learning health and rehabilitation features of a new registered user in an aggregated manner from multiple clients, and updating the prediction model of each client.

2. The system of claim 1, wherein, The client specifically comprises: a feature monitoring and collecting module, for collecting physiological feature data and posture feature data; a data analysis module, for predicting corresponding symptoms by using a prediction model according to the physiological feature data and the posture feature data, and matching health and rehabilitation equipment parameters corresponding to the symptoms in combination with a knowledge graph.

3. The system of claim 2, wherein, The feature monitoring and collecting module specifically comprises: a physiological feature data collecting submodule, for collecting heart rate, blood pressure and blood oxygen data of a user by using a flexible sensor to obtain physiological feature data; a posture feature data collecting submodule, for acquiring posture feature data of the user by using a millimeter wave radar.

4. The system of claim 3, wherein, The feature monitoring and collecting module further comprises: a data preprocessing module, for preprocessing the collected data based on a sensor space-time alignment algorithm.

5. The system of claim 2, wherein, The data analysis module specifically comprises: a model construction submodule, for constructing a prediction model, wherein the prediction model comprises a feature extraction layer and a personalized prediction layer connected in sequence, the backbone network of the feature extraction module is a ResNet-18 network, the auxiliary network is a BiLSTM network, and the personalized prediction layer is composed of a full connection layer and an activation function; a symptom prediction submodule, for inputting the physiological feature data and the posture feature data into the prediction model for classification prediction to obtain a symptom prediction result; a knowledge graph matching submodule, for matching a knowledge graph according to the symptom prediction result to generate device control parameters; the knowledge graph stores a symptom-device parameter mapping rule.

6. The system of claim 1, wherein, The distributed optimization unit specifically comprises: an initialization module, for initializing a federated learning training environment through a central server, generating an initial prediction model and setting a training round regulation value; a pre-training module, for pre-training each module in the initial prediction model by using the central server, and distributing the pre-trained modules to each client; a client training module, for training each pre-trained module through the client to obtain an updated feature extraction module and an updated personalized module; the client separately trains the updated personalized module to obtain a separately trained personalized module and save it in the client; a feature extraction module aggregation module, for sending the updated feature extraction module to the central server for weighted average calculation; the central server judges the current training round, and if the preset round is reached, the training is ended, the weighted average calculated feature extraction module is sent to the client, the client stops training, and the client uses the separately trained personalized module and the weighted average calculated feature extraction module for symptom prediction.

7. The system of claim 6, wherein, The pre-training module specifically comprises: A dataset construction module is configured to create a medical dataset by using a generative adversarial network; A training module is configured to input the medical dataset into an initial prediction model to perform classification prediction, and train the initial prediction model according to a target loss function to obtain a pre-trained prediction model.

8. The system of claim 6, wherein, The client training module specifically includes: A data acquisition module is configured to acquire local data of the client; The training module is configured to train the pre-trained modules according to the local data to obtain an updated feature extraction module and an updated individualization module; the updated individualization module is separately trained based on small sample meta learning to obtain a separately trained individualization module and save the separately trained individualization module in the client.

Images