CN112309581A

CN112309581A - Private medical intelligent auxiliary method based on machine learning

Info

Publication number: CN112309581A
Application number: CN202011208327.XA
Authority: CN
Inventors: 孟祥福; 薛琪; 张霄雁; 朱尧; 温晶; 李政
Original assignee: Liaoning Technical University
Current assignee: Liaoning Technical University
Priority date: 2020-11-03
Filing date: 2020-11-03
Publication date: 2021-02-02

Abstract

The invention discloses a private medical intelligent auxiliary method based on machine learning, which comprises the following steps: detecting a heart rate value and a blood oxygen value; detecting body temperature; simple motion state detection; realizing a link network and a positioning service; binding a user; analyzing the health state; monitoring of genetic diseases; monitoring of infectious diseases; and (5) monitoring emergent infectious diseases. According to the private medical intelligent auxiliary method based on machine learning, disclosed by the invention, a private medical intelligent auxiliary system based on machine learning is constructed by integrating a plurality of novel technologies such as monitoring information, a wireless communication technology, a cloud platform database, a knowledge map, data visualization and the like of an intelligent terminal, so that the problem of insufficient medical resources can be favorably alleviated, the science popularization can be effectively realized, infectious diseases can be effectively prevented, certain deep excavation and prediction work can be performed by combining big data, and the establishment and development of basic medical health informatization are accelerated.

Description

Private medical intelligent auxiliary method based on machine learning

Technical Field

The invention belongs to the technical field of medical treatment and health, and particularly relates to a private medical intelligent auxiliary method based on machine learning.

Background

The medical health cause relationship is closely related to the personal interests of people, and is a hot spot of high social concern. Market research results show that people want to obtain various index data of their bodies in daily life, more hope that they know the etiology and how to treat quickly when they are ill, and hope to master real-time information about epidemic situation conditions in a major epidemic period. The intelligent medical service is a service system integrating cloud computing, the Internet of things and patient data, with the development and the improvement of network big data, a complete health service management platform is provided for the personal health of a patient by a model of the intelligent medical service, and the problems that the application function is single, a unified and efficient intelligent medical network system is not formed and the like exist.

Disclosure of Invention

Based on the defects of the prior art, the technical problem to be solved by the invention is to provide a private medical intelligent auxiliary method based on machine learning, which can help to alleviate the problem of insufficient medical resources, can effectively remove science popularization and prevent infectious diseases, and can be combined with big data to do some deep mining and prediction work, so that the establishment and development of basic medical health informatization are accelerated.

In order to solve the technical problem, the invention provides a private medical intelligence auxiliary method based on machine learning, which comprises the following steps:

step 1: detecting a heart rate value and a blood oxygen value;

step 2: detecting body temperature;

and step 3: simple motion state detection;

and 4, step 4: realizing a link network and a positioning service;

and 5: binding a user;

step 6: analyzing the health state;

and 7: monitoring of genetic diseases;

and 8: monitoring of infectious diseases;

and step 9: and (5) monitoring emergent infectious diseases.

Optionally, in step 1, the detection of the heart rate value and the blood oxygen value is realized by a module of a biosensor integrating a pulse oximeter and a heart rate monitor.

Optionally, in step 2, the body temperature is detected by a digital temperature sensor with a model MAX 30102.

Further, in step 6, the analysis of the health status generates a health report of the genetic disease and an infectious disease report according to the disease condition of the infectious disease and the family genetic history, the database keyword query is performed through the data in the database by using a method based on a data map, and the data map is directly processed, so that the screening is performed, and the special popularization protection content of the family genetic disease department is generated and pushed to the user.

Optionally, in the step 7 of monitoring the genetic disease, cases uploaded by the user are classified according to age and name attributes, a family genetic disease knowledge map is generated according to the cases through a data analysis technology, and after the user is monitored to suffer from a certain genetic disease, the system automatically records data before and after treatment to judge whether the disease condition is improved.

Optionally, in the step 8 of monitoring infectious diseases, the common infectious disease occurrence conditions are obtained through Python language, and are visualized through charts such as histograms, and the occurrence time of the infectious diseases is reminded according to the prediction result.

According to the private medical intelligent auxiliary method based on machine learning, disclosed by the invention, a private medical intelligent auxiliary system based on machine learning is constructed by integrating multiple novel technologies such as monitoring information of an intelligent terminal, a wireless communication technology, a cloud platform database, a knowledge map, data visualization and the like. In the hardware manufacturing, the range measurement work between the tag and the hardware architecture design scheme is completed by using a communication interface fusing MAX30102, DS18B20, MPU-6000(6050), ARM + Ethernet + CAN +1MSF + Wifi and using an SDS-TWR algorithm to obtain the positioning data. The science popularization content is pushed by using a keyword query technology based on a data graph method, and the system carries out classification identification on cases by using named entity identification in deep learning. Adopting a better bert model than word2vec to perform transfer learning, performing model pre-training on fifty percent of selected cases, and training the cases by using bert + crf after training to form attributes such as names, ages, disease names, disease histories and the like. The knowledge map of the family genetic diseases is generated by using neo4j software in the form of entity-relation-attribute triples. And predicting the time and level of the next stage of the infectious disease by using a time series model. For the serious emergent infectious diseases which occur in the same year, a large amount of data is collected by the system to carry out epidemic prevention and control. The epidemic situation information is crawled in real time by using a crawler technology in Python, the epidemic situation information is updated every five minutes, data is represented in real time by using a histogram, a histogram and the like by using an AJAX technology, the data displayed to a user is refreshed every five minutes, and a thermodynamic diagram is drawn by using a javascript API interface of a Gade map.

The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical means of the present invention more clearly understood, the present invention may be implemented in accordance with the content of the description, and in order to make the above and other objects, features, and advantages of the present invention more clearly understood, the following detailed description is given in conjunction with the preferred embodiments, together with the accompanying drawings.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly described below.

FIG. 1 is a functional diagram of a machine learning-based personal medical intelligent assistance method of the present invention;

FIG. 2 is a hardware flow diagram of the present invention;

FIG. 3 is a radar chart of basic indicators of a human body;

FIG. 4 is a life-related diagram of a person's physical condition;

FIG. 5 is a flow chart of the Bert Model;

FIG. 6 is a family disease history knowledgegraph presentation;

fig. 7 is a diagram showing an overall configuration of a prediction model.

Detailed Description

Other aspects, features and advantages of the present invention will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, which illustrate, by way of example, the principles of the invention and, together with the description, further serve to explain the principles of the invention. In the referenced drawings, like or similar elements in different drawings are designated with the same reference numerals.

Aiming at the problems that a doctor cannot know the physical signs of a patient in a period of time to make a definite diagnosis, a remote area cannot see a patient due to unreasonable hospital configuration and a medical worker lacks to make a long queuing time, the invention provides a private medical intelligent auxiliary system based on machine learning, which comprises a cloud server, a client and hardware facilities, and is shown in figure 1.

Firstly, hardware: the hardware part comprises private medical treatment wisdom auxiliary system's test terminal, wears in user's wrist position, can realize the function of physiological parameters such as monitoring blood oxygen value, heart rate value, body temperature, simple and easy motion state, gathers user's physiological parameter data package, provides the basis for big data analysis and suggestion propelling movement. The hardware flow chart is shown in fig. 2, and the specific method is as follows:

step 1, hardware realization of heart rate value and blood oxygen value detection function

Heart rate value, blood oxygen value detect function: MAX30102 is a module of an integrated pulse oximeter and heart rate monitor biosensor. It integrates a red LED and an infrared LED, a photodetector, optics, and low noise electronic circuitry with ambient light rejection. MAX30102 adopts a 1.8V power and an independent 5.0V power that is used for inside LED, is applied to wearable equipment and carries out heart rate and blood oxygen collection and detect, wears in places such as finger, earlobe and wrist. The standard I2C compatible communication interface can transmit the collected values to the single chip microcomputer for heart rate and blood oxygen calculation.

Step 2, hardware realization of body temperature detection function

Body temperature detection function: DS18B20 is a commonly used digital temperature sensor, and the output of it is a digital signal, has small, the hardware cost is low, and the interference killing feature is strong, the high-accuracy characteristics. The DS18B20 digital temperature sensor is convenient to wire and can be applied to various occasions after being packaged, and the excellent anti-interference capability and the simple hardware circuit connection make the sensor a better choice for the system construction.

Step 3, hardware implementation of simple motion state detection function

Simple motion state detection function: MPU-6000(6050) is the first global integrated 6-axis motion processing component, compared with a multi-component scheme, the problem of time axis difference between a combined gyroscope and an accelerator is solved, a large amount of volume is reduced, the rotation angular speed during deflection and inclination is measured more accurately, and the current posture of a user can be obtained through calculation.

Step 4, realizing the link network and the positioning service

In the aspect of linking networks, a hardware architecture design scheme of a communication interface fused by ARM + Ethernet + CAN +1MSF + Wifi is adopted, a flexible configuration protocol is designed by self, fusion of various industrial communication interfaces and an optical network is realized, system efficiency is accelerated by optimizing software architecture design, and driver design of a hardware system is completed.

In the positioning service, the SDS-TWR algorithm is used for completing the distance measurement work between the positioning service and the label, acquiring positioning data and completing the forwarding and uploading of the positioning data.

And secondly, building a web project.

In order to facilitate a user to check real-time and recent vital signs at any time, the project is constructed by Python + django + mysql, a front-end page adopts a grid system of borestrap, so that the interface is more attractive, and the project is suitable for screens of various sizes, thereby building a Web application system.

The specific method is as follows:

step 5, binding the user

(1) User registration

Firstly, the user registers user data on the software, and the ASP is used for verifying whether the user information is effective or not. After success, the information of name, age, height, weight and the like is input and then is stored in a background database.

(2) User information real-time update

And reading the data of the user from the server. The system generates a dynamic graph by applying Basic Radar Chart of Echarts according to the detected heart rate, body temperature, blood pressure, blood oxygen and other data of the wearer. Fig. 3 is a radar chart showing basic indexes of a human body. When the value is abnormal and the specified time is exceeded, the system can remind the wearer through a hand ring. When the user is in danger, the alarm system can be enabled to be effective.

Utilize Lora location technique to make the bracelet learn self position, the event waits for all sensors to gather data, uploads the server with data after all flag bit positions promptly. CPU resources are saved, the same data are prevented from being uploaded for multiple times, and the efficiency is improved.

Step 6, health status analysis function

(1) Generating personal health reports

The system can generate a health report of the genetic disease and an infectious disease report according to the incidence condition of the infectious disease and the family genetic history. The report is predicted in real time and continuously changed according to the society and the condition of the wearer, and the user can inquire about the related content in real time according to the report to know.

(2) Pushing science popularization content using keyword query technology

Performing database keyword query on data in a database by using a method based on a data graph, directly processing the data graph, enumerating simplified subtrees from the data graph, firstly, regarding the data graph as a weighted data graph, and assuming that the data graph is materialized; secondly, the weights of nodes (tuples) and edges (main external key association between tuples) in the data graph are used as a keyword query to find top-k simplified subtrees with the minimum cost. Therefore, screening is carried out, and exclusive family genetic disease department popularization and protection contents are generated and pushed to users. Fig. 4 shows a daily life knowledge of the physical condition of a person.

Step 7, genetic disease monitoring function

After the user manually uploads the cases, the system carries out classified identification on the cases by utilizing named entity identification in deep learning. And adopting a better bert model than word2vec for transfer learning. The Model flow is as in the Bert Model flow chart of FIG. 5. Fifty percent of cases are selected to be used as model pre-training, and the cases are trained by using bert + crf after training is finished, so that the attributes of name, age, disease name, disease history and the like are formed. The model adopts a traditional IOB labeling mode and marks each character with a label.

After successful modeling, the well-learned CRF model is used for new observation sequence (O)₁,O₂,O₃....O_i) Finding out a most probable hidden state sequence i₁,i₂...i_iAnd the path solving process adopts a viterbi algorithm. Finally, the processed case information is summarized and sorted.

The knowledge graph is a mapping map of the knowledge field, and the potential relation effect among entities is obviously mined for expressing the relation among the entities, so that the knowledge graph of the family genetic disease is generated by using the neo4j software in the form of entity-relation-attribute triples according to the attributes and by taking the name of a person as a central node. The data storage manner as shown in fig. 6 is: { "name": "clear", "relation": "suffering from", "disease": "Heart disease" } genetic disease prediction section used XGBost + LightGBM + LSTM. The first two classes can be regarded as tree models, and the LSTM is a neural network model. The two models have larger principle difference, the generated result has lower correlation, and the fusion is favorable for improving the prediction accuracy. The specific model structure is shown in fig. 7.

The prediction steps are as follows:

1. and loading data, and learning by using the trained model.

2. Data is loaded and variables X (various vital sign data) and label Y (diabetes mellitus is acquired) are separated.

3. The data set is divided into a training set and a testing set, the training set is used for training the model, and the testing set is used for testing the accuracy of the model.

4. XGBoost packaged classifiers and regressors are used, and XGBPassifiers can be directly used for establishing a model.

Predicting the overall structure of the model:

the XGBoost _1 is used to learn the feature combination F1 to obtain the prediction result of XGBoost _1 (including the prediction results for the training set and the test set), which is added as a new feature to the feature combinations F2 and F3 as the input features of the second layer LightGBM _1 and LightGBM _2, respectively, the result of LightGBM _1 is added as a new feature to the feature combination F4 as the input feature of the third layer XGBoost _2, and the third layer contains an LSTM model, which is trained by using the feature combination F5, and the result of the second layer LightGBM _2 is weighted and fused with the prediction results of the third layer XGBoost _2 and LSTM as the final result.

Adjusting parameters:

the following are the general practical best values of the three hyper-parameters, which can be set to this range first, then left curves are drawn, and then the parameters are reconciled to find the best model:

the learning _ rate is 0.1 or less, and the smaller the learning _ rate, the more weak the learner needs to be added; 2-8 of tree _ depth; subsample is 30% -80% of the training set;

the adjustable hyper-parameter combinations are:

the number and size of the trees (n _ estimators and max _ depth).

Learning rate and number of trees (learning _ rates and n _ estimators).

Subsampling rates of ranks (subsample, subsample _ byte and subsample _ byte).

Step 8, infectious disease monitoring function

The infectious disease monitoring mainly aims at predicting the onset time of infectious diseases, obtaining the onset conditions of common infectious diseases through Python language, visualizing the onset conditions through a histogram chart, and reminding the onset time of the infectious diseases according to the prediction result.

Step 9, infectious disease monitoring function

For the serious emergent infectious diseases which occur in the same year, a large amount of data is collected by the system to carry out epidemic prevention and control. The method has the advantages that epidemic situation information is crawled in real time by using a crawler technology in Python and is updated every five minutes, and AJAX has the greatest advantage that data can be exchanged with a server and partial webpage content can be updated under the condition that the whole page is not reloaded. The data is represented in real time by histograms, bar charts, etc. using ajax technology, and the data presented to the user is refreshed every five minutes. The javascript api interface of the high-grade map is utilized.

While the foregoing is directed to the preferred embodiment of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.

Claims

1. A private medical intelligent assisting method based on machine learning is characterized by comprising the following steps:

step 1: detecting a heart rate value and a blood oxygen value;

step 2: detecting body temperature;

and step 3: simple motion state detection;

and 4, step 4: realizing a link network and a positioning service;

and 5: binding a user;

step 6: analyzing the health state;

and 7: monitoring of genetic diseases;

and 8: monitoring of infectious diseases;

and step 9: and (5) monitoring emergent infectious diseases.

2. Private medical intelligent assistance method based on machine learning according to claim 1, characterised in that in step 1 the detection of heart rate values, blood oxygen values is carried out by means of a module integrating a pulse oximeter and a biosensor of a heart rate monitor.

3. The personal medical intelligent assistance method based on machine learning as claimed in claim 1, wherein in step 2, the detection of the body temperature is realized by a digital temperature sensor with model MAX 30102.

4. The private medical intelligent assistance method based on machine learning as claimed in claim 1, wherein in step 6, the analysis of health status generates a health report of genetic disease and an infectious disease report according to the disease condition of infectious disease and family genetic history, and the data in the database is used for database keyword query by using a method based on a data map, and the data map is directly processed, so as to perform screening and generate a proprietary family genetic disease popularization protection content to be pushed to the user.

5. The personal medical intelligent assistance method based on machine learning as claimed in claim 1, wherein in the step 7 of monitoring the genetic disease, cases uploaded by the user are classified according to age and name attributes, a family genetic disease knowledge map is generated according to the cases by a data analysis technique, and after a certain genetic disease is monitored, the system automatically records data before and after treatment to judge whether the disease condition is improved.

6. The private medical intelligent assistance method based on machine learning according to claim 1, wherein in the infectious disease monitoring of step 8, the common infectious disease onset is obtained in Python language and visualized by histogram chart, and the time of infectious disease onset is reminded according to the prediction result.