An intelligent risk assessment system and method for pediatric patients' condition

Abstract

The present invention discloses an intelligent risk assessment system and method for the condition of a child patient. First, data related to the condition is collected from the family members and the hospital of the child patient; then, the child patient representation related to the condition is extracted from the designed data collection table to construct entity nodes, and the relationship between entities is constructed according to the expert knowledge and the medical knowledge base to construct a knowledge graph of the risk factors of the condition of the child patient; then, a dynamic attribute graph of the condition of the child patient at multiple moments is constructed based on the collected data and the knowledge graph of the risk factors of the condition of the child patient, and the attribute graph is learned using a dynamic heterogeneous graph neural network model to output the predicted values ​​of important representations, and the degree of criticality is evaluated based on the predicted values ​​of important representations, and the treatment strategy of the child patient is recommended. The present invention uses historical data and current data to ensure the accuracy of the assessment of the child patient, and focuses on the analysis and prediction of the representation of the child patient, avoiding the risk of misdiagnosis and alleviating the ethical issues of smart medical care.

Description

Technical Field

[0001] The present invention relates to the field of child care, and in particular to an intelligent disease risk assessment system and method for child patients. Background Art

[0002] Since pediatric diseases have the characteristics of rapid onset, rapid development, and poor regulatory ability, family members often do not pay attention to the condition of children at the beginning of the illness. In addition, the medical process takes a long time and is cumbersome, which further hinders family members from taking children to see a doctor when the symptoms are not serious. On the other hand, some family members are too sensitive to the condition of children and occupy medical resources when medical treatment is not needed. Therefore, it is very important to provide children with preliminary critical illness warnings and medical decision-making recommendations before medical treatment. This can alleviate the uneven distribution of medical resources, optimize the integration of medical data, and make the medical industry precise, convenient, and intelligent.

[0003] The vigorous development of big data technology and communication technology has provided new ideas for the above problems. It can form a medical service model centered on patient data and realize the interactive model between patients and medical staff, medical institutions, etc., that is, smart medical care. On the one hand, driven by the development of big data technology, the massive data generated by the hospital information system can be efficiently utilized through data processing, data mining, etc., combined with medical big data mining, artificial intelligence technology, etc., to achieve intelligent diagnosis and medical decision-making recommendations for patients. On the other hand, the high speed, low latency, and high reliability of 5G have promoted the construction of smart medical care, which can ensure efficient interaction between child patients and medical institutions.

[0004] However, smart healthcare faces challenges in medical ethics. Due to the uncertainty of smart healthcare, including the uncertainty of data analysis and artificial intelligence, there is a deviation between the algorithm results and the actual situation, causing physical and financial losses to patients. If smart healthcare misdiagnoses, there will be disputes over who will be held responsible for medical and legal responsibilities, and it may undermine the balance between humanity and technology. Human life data is very complex, and even professional doctors can make misdiagnoses, not to mention artificial intelligence, which has not yet been fully industrialized and mature. This has led to many people questioning smart healthcare, which has seriously hindered its development.

[0005] On the other hand, most of the applications of smart medical care are for the general public, that is, people of all ages, and there are fewer studies on children. However, compared with adults, children do not have the ability to seek medical treatment independently and have poor expression skills, which can easily cause family members to ignore the condition of children and delay medical treatment. Therefore, children need more critical illness warnings than adults, and they need more intelligent technologies for information acquisition, information fusion, information mining, and historical information utilization of children's diseases. In order to ensure that when children cannot accurately describe their own symptoms like adults and answer the doctor's questions accordingly, the degree of critical illness of children can be assessed through data mining and analysis, which will attract the attention of family members and doctors. Summary of the invention

[0006] Purpose of the invention: Based on the importance of pediatric smart healthcare and the lack of research, as well as the ethical issues of smart healthcare, the purpose of the present invention is to provide an intelligent risk assessment system and method for pediatric patients' conditions, which can assess the criticality of pediatric patients and provide medical decision-making recommendations by collecting, analyzing and mining pediatric patient data.

[0007] Technical solution: To achieve the above-mentioned invention purpose, the present invention provides an intelligent risk assessment system for children's condition, comprising:

[0008] Data collection module, medical characteristics analysis module for children, and condition prediction and decision-making module for children;

[0009] The data collection module is used to collect data related to the condition from the family members of the child patient and the hospital, store it after pre-processing, and support the call of the child patient's historical data;

[0010] The child patient condition representation data analysis module is used to extract child patient representations related to the condition from the designed data collection table to construct entity nodes, build relationships between entities based on expert knowledge and the medical knowledge base, and build a knowledge graph of child patient condition risk factors; and calculate the association between entities to verify and improve the relationship between entities;

[0011] The child patient's condition prediction and decision-making module is used to construct a dynamic attribute graph of the child patient's condition at multiple moments based on the collected data and the knowledge graph of the child patient's condition risk factors, and then use the dynamic heterogeneous graph neural network model to learn the attribute graph, output the predicted values ​​of important representations, evaluate the degree of criticality based on the predicted values ​​of the important representations, and make recommendations on the child patient's treatment strategy; wherein the adjacency matrix of the attribute graph input into the dynamic heterogeneous graph neural network model is determined according to the topological structure of the knowledge graph provided by the child patient's condition representation data analysis module, each relationship type in the knowledge graph corresponds to an adjacency matrix, and the feature matrix is ​​determined according to the data collected at each moment by the data collection module; the important representation is the screened-out condition representation whose importance to the condition is greater than a set threshold.

[0012] Optionally, the data collection module includes a data collection unit for the family of the child patient, a data collection unit for the hospital, a historical data storage and call unit, and a data preprocessing unit;

[0013] The data collection unit for the family members of the child patient is used to collect the disease-related data provided by the family members, including the demographic characteristics of the child patient, the disease condition, the hospital visited, and the physical signs that can be measured by the family;

[0014] The hospital data collection unit is used to collect the disease-related data provided by the hospital, including the objective and subjective physical manifestations of the child patient, which are divided into in vitro test data and body fluid test data;

[0015] The objective in vitro test is to test the surface of the child patient's body with an instrument, the subjective in vitro test is based on the attending physician's assessment of the child patient's physical signs, and the objective body fluid test is based on drawing blood or other body fluids for testing;

[0016] The historical data storage and calling unit is used to establish an index in the database based on the child patient identity information, and store the child patient information at a location corresponding to the index according to the time of collection, so as to form a data table of the child patient data time series with the child patient identity information as the index;

[0017] The data preprocessing unit is used to perform data cleaning, data conversion and data filling on the collected data.

[0018] Optionally, the pediatric patient condition representation data analysis module includes a condition risk factor knowledge graph construction unit, a knowledge graph deep analysis unit, and a knowledge graph deep fusion unit;

[0019] The disease risk factor knowledge graph construction unit is used to extract entity nodes from the child patient family data collection table and the hospital data collection table and establish edge relationships between the nodes to construct a knowledge graph; the extracted entity nodes are the names of physical manifestations that affect the child patient's condition or may bring risks to the child patient's physical health from the collection table, and these manifestations are divided into four types of nodes according to medical classification: cardiovascular / nervous system life manifestations, acid / base manifestations, biochemical detection manifestations, and blood manifestations;

[0020] The node edge relationship is established based on expert knowledge and a medical knowledge base, the mutual influence of the extracted body representations is modeled, and nodes that influence each other or represent the same condition are connected;

[0021] The knowledge graph in-depth analysis unit is used to calculate the degree of association between entity nodes based on the medical big data of pediatric patients stored in the database using an association analysis algorithm, determine the existence of edges in the knowledge graph, and delete edges with a degree of association lower than a set threshold;

[0022] The knowledge graph deep fusion unit is used to update the knowledge graph, including data fusion and knowledge fusion; the data fusion refers to updating the number and / or type of nodes in the knowledge graph of child patient condition risk factors based on the updated child patient family data collection form and hospital data collection form; the knowledge fusion refers to updating the number and / or type of edges in the knowledge graph of child patient condition risk factors based on the updated expert knowledge and medical knowledge base.

[0023] Optionally, the child patient condition prediction and decision-making module includes a child patient condition dynamic attribute graph construction unit, a child patient condition prediction model unit and a child patient condition assessment unit;

[0024] The child patient's condition dynamic attribute graph construction unit is used to construct a child patient's condition dynamic attribute graph, including the child patient's condition representation at the current moment and the child patient's condition representation at previous moments; wherein the child patient's condition representation attribute graph at the current moment is established based on the child patient's condition risk factor knowledge graph, wherein the attribute value includes the specific value or severity description of the corresponding representation; the attribute value in the child patient's condition representation attribute graph at previous moments is provided by the child patient's historical data, and is stacked and merged with the child patient's condition representation attribute graph at the current moment in a time series to form a dynamic graph;

[0025] The child patient condition prediction model unit is used to input the child patient condition dynamic attribute graph into the child patient condition intelligent diagnosis model to predict the value of the child patient's important representation, wherein the child patient condition intelligent diagnosis model is a dynamic heterogeneous graph neural network model trained with historical data provided by the data acquisition module;

[0026] The important representations of the child patients are screened out according to a feature selection model based on machine learning. The feature selection model is trained through historical data of child patients, with the criticality of the child patients as the dependent variable and the representations of the child patients as the independent variable. The relationship between the independent variable and the dependent variable is modeled, and parameter values ​​are assigned to the independent variables. Finally, the importance of the independent variable is measured by the size of the parameter value.

[0027] The child patient condition assessment unit is used to score the criticality of the child patient based on the predicted values ​​of the child patient's important characteristics and in combination with the child critical patient scoring table, and give immediate medical treatment or self-selected medical treatment advice based on the score; if immediate medical treatment is recommended, the map interface will be connected to find nearby hospitals and route traffic conditions and recommend hospitals based on the objective conditions of the child patient.

[0028] Among them, the objective conditions of child patients are provided by the data collection form for the family members of child patients, including whether they are currently receiving treatment in the hospital, the location of the child patients, and whether there are conditions for transporting the child patients, etc.;

[0029] Optionally, the child patient condition assessment unit is also used to send the predicted values ​​of the child patient's important manifestations and the collected data to the doctor of the hospital selected by the family for remote diagnosis, with the consent of the family and without violating the privacy of the child patient.

[0030] The present invention provides a method for intelligent risk assessment of a pediatric patient's condition, comprising the following steps:

[0031] Collecting data related to the condition from the family members of the child patients and the hospital, storing them after pre-processing, and the stored data includes historical data of the child patients collected at different times;

[0032] Extract the representation of children's patients related to their conditions from the designed data collection table to construct entity nodes, build the relationship between entities based on expert knowledge and the medical knowledge base, and build a knowledge graph of risk factors for children's conditions; and calculate the correlation between entities to verify and improve the relationship between entities;

[0033] Based on the collected data and the knowledge graph of risk factors of children's conditions, a dynamic attribute graph of the children's condition at multiple moments is constructed, and then the attribute graph is learned using a dynamic heterogeneous graph neural network model to output predicted values ​​of important representations. The degree of criticality is evaluated based on the predicted values ​​of the important representations, and treatment strategies for children are recommended; wherein the adjacency matrix of the attribute graph input into the dynamic heterogeneous graph neural network model is determined according to the topological structure of the knowledge graph provided by the children's condition representation data analysis module, each relationship type in the knowledge graph corresponds to an adjacency matrix, and the feature matrix is ​​determined according to the data collected at each moment by the data collection module; the important representations are the screened-out condition representations whose importance to the condition is greater than a set threshold.

[0034] The present invention provides a computer system comprising a memory, a processor and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the steps of the method for intelligent risk assessment of a child's condition are implemented.

[0035] Beneficial effects: The intelligent risk assessment system and method for the condition of children provided by the present invention, on the one hand, uses data mining and knowledge graph technology to propose a novel method for analyzing data of children, which can effectively and dynamically combine medical big data and medical knowledge base, model the relationship between data, and mine new knowledge through association analysis. The knowledge graph of risk factors of children's condition constructed not only helps medical personnel understand and analyze data in a visual form, but also integrates unstructured data into structured graph data for intelligent prediction models. In the intelligent prediction of the condition of children, the present invention comprehensively uses historical data and current data, avoids the influence of the vague description of the current condition of the children, and ensures the accuracy of the assessment of the children. On the other hand, the present invention also provides a method for avoiding the risk of misdiagnosis and alleviating the ethical issues of smart medical care. Due to the high complexity of human life data and the uncertainty of artificial intelligence, smart medical care is prone to misdiagnosis and causes social problems. The present invention focuses on analyzing and predicting the patient's representation, and does not directly make a difficult clinical diagnosis, but makes a critical illness warning for the patient based on the predicted value of the important representation and sends the patient's data to the doctor for remote diagnosis. This not only avoids the risk of misdiagnosis and alleviates the ethical issues of smart medical care, but also uses artificial intelligence to mine information in the data to help child patients receive timely medical treatment, alleviate the shortage of medical resources and assist doctors in diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a schematic diagram of the architecture of the intelligent risk assessment system for pediatric patients' condition provided by an embodiment of the present invention;

[0038] Figure 2 is a flow chart of a method for collecting data of pediatric patients in an embodiment of the present invention;

[0039] Figure 3 is a schematic flow chart of a method for analyzing medical characteristics of pediatric patients in an embodiment of the present invention;

[0040] Figure 4 is a flow chart of a method for predicting and making decisions on the condition of a child patient in an embodiment of the present invention;

[0041] Figure 5 It is a flow chart of the intelligent risk assessment method for the condition of a child patient provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] like Figure 1 As shown, an intelligent risk assessment system for a child patient's condition disclosed in an embodiment of the present invention mainly includes: a data acquisition module, a child patient's medical feature analysis module, and a child patient's condition prediction and decision-making module; wherein,

[0044] (1) A data collection module, which is used to collect and pre-process the data related to the disease from the child patient data provided by the child patient's family and the hospital, and to reasonably store and call the child patient's historical data; wherein the child patient data provided by the child patient's family is based on the designed child patient family data collection form to collect the child patient's demographic characteristics, disease conditions, and simple child patient physical signs that can be measured by the hospital and the family; the child patient data provided by the hospital is based on the designed hospital data collection form to collect the child patient's objective and subjective physical manifestations, which are divided into in vitro testing and body fluid testing; the storage and calling of the child patient's historical data is used to store the collected child patient data in the background database according to the collection time to form a large medical database of the child patient, and to call the child patient's historical data to assist in the diagnosis of the child patient;

[0045] Optionally, in an embodiment of the present invention, the child patient data collection module at least includes a child patient family data collection unit, a hospital data collection unit, a data preprocessing unit, and a historical data storage and calling unit;

[0046] The data collected by the child patient's family data collection unit specifically include the child patient's demographic characteristics, the hospital where the patient is being treated, the condition of the disease, and simple vital signs of the child patient. Among them, the child patient's demographic characteristics include age, gender, address, etc.; the hospital where the patient is being treated includes whether the child patient is currently being treated in the hospital, the hospital grade, the region, etc.; the condition of the disease includes the diagnosis upon admission, the department where the child patient is admitted, whether the child patient is postoperative, etc.; simple vital signs of the child patient include body temperature, heart rate, whether the limbs are cold, etc. If the family member enters yes for the data on whether the child patient is currently being treated in the hospital, then the hospital data collection is required, otherwise it is not required;

[0047] The data collected by the hospital data collection unit specifically include objective in vitro tests, subjective in vitro tests, and objective body fluid tests of child patients, wherein objective in vitro tests refer to tests on the body surface of child patients using instruments, such as systolic blood pressure, diastolic blood pressure, and respiration, etc.; subjective in vitro tests are based on the attending physician's assessment of the child's physical signs, such as cyanosis, capillary refill time, and pupil light reflex, etc.; objective body fluid tests are based on blood draws or body fluids from other parts for testing, such as blood sugar, base excess, and calcium ion concentration, etc.;

[0048] The data preprocessing unit processes the incomplete, inconsistent and erroneous data in the raw data collected by the above-mentioned unit, including data cleaning, conversion and filling, to alleviate the impact of these abnormalities on the condition assessment method of pediatric patients, and convert the raw data into a form that is easy for data mining;

[0049] The historical data storage and calling unit is used to store the data of all child patients and can call out the data of a certain child patient at all times. The storage function encodes the child patient's identity information into a hash code as an index and stores it in the background SQL database with the collected moments as a sequence. The calling function encodes the child patient's identity information into a hash code and uses this as an index to query the database for the data of the child patient at all times.

[0050] Specifically, the intelligent risk assessment framework for children's conditions also includes:

[0051] (2) Child patient condition representation data analysis module: This module is based on knowledge graph technology and combines the relationship between the collected child patient representation data and medical expert knowledge analysis data to establish a knowledge graph of child patient condition risk factors. It also conducts in-depth analysis and fusion of the knowledge graph based on the child patient data and the medical knowledge base.

[0052] Optionally, in an embodiment of the present invention, the child patient condition representation data analysis module includes at least a condition risk factor knowledge graph construction unit, a knowledge graph deep analysis unit and a knowledge graph deep fusion unit;

[0053] The disease risk factor knowledge graph construction unit constructs a knowledge graph based on child patient data and medical expert knowledge using a knowledge graph construction method. The knowledge graph is composed of factors related to the disease and triplets formed by the influence relationship between these factors based on medical expert knowledge. The factors related to the disease are based on representation names related to the child patient's disease extracted from data collected from the child patient's family and the hospital; the influence relationship between the factors refers to the correlation between medical representations, where multiple representations have an impact on the same aspect of the child patient's body or one representation has an impact on multiple aspects of the child patient's body;

[0054] The knowledge graph deep analysis unit is based on the collected medical big database of pediatric patients, and uses the association algorithm to verify the constructed knowledge graph of pediatric patients' disease risk factors, determine the existence of each node and edge relationship in the knowledge graph, and improve the association relationship between each entity. The medical big database of pediatric patients is the data of all pediatric patients stored in the background by the historical data storage and call unit; the association algorithm refers to calculating the Pearson correlation coefficient between nodes as their correlation based on the medical big data of pediatric patients; based on the correlation between nodes and the relationship threshold, check whether the edges between nodes are correct, and delete them if they are lower than the set threshold;

[0055] The knowledge graph deep fusion unit updates the nodes and edges of the knowledge graph, including data fusion and knowledge fusion. Data fusion is based on the updated child patient data collection table to update the number and / or type of nodes in the child patient disease risk factor knowledge graph; knowledge fusion is based on the increasingly rich professional knowledge sources to integrate knowledge systems from different sources with differences, and based on the updated expert knowledge and medical knowledge base, to update the number and / or type of edges in the child patient disease risk factor knowledge graph.

[0056] Specifically, the intelligent risk assessment framework for children's conditions also includes:

[0057] (3) The child patient's condition prediction and decision-making module builds a dynamic attribute graph of the child patient's condition at multiple moments based on the collected child patient data and the knowledge graph of the child patient's condition risk factors. The module uses the child patient's condition intelligent diagnosis model to learn the attribute graph and output the predicted values ​​of important representations. The module evaluates the criticality of the child patient's condition based on the predicted values ​​of important representations, and makes recommendations on the child patient's medical treatment based on the results. The module also integrates the child patient's data and sends it to the doctor for remote diagnosis.

[0058] The dynamic attribute graph of the child patient's condition is composed of a combination of a child patient's condition representation graph at the current moment and a child patient's condition representation graph at previous moments. The child patient's condition representation graph at the current moment is established based on the child patient's condition risk factor knowledge graph, wherein the attribute value includes the specific value or severity description of the corresponding representation; wherein the attribute value in the child patient's condition representation attribute graph at previous moments is provided by the historical data storage and calling unit, and the child patient's condition representation at all moments recorded in recent time is called to construct an attribute graph, and the time series is stacked and merged with the child patient's condition representation attribute graph at the current moment into a dynamic graph;

[0059] The intelligent diagnosis model for children's conditions is a prediction model for children's patient representations based on a dynamic heterogeneous graph neural network. It takes a dynamic attribute graph of children's conditions as input, and uses the current and past data of children to predict the predicted values ​​of important representations of children in the future. The prediction of important representations is based on the important signs that have a great impact on the critical condition of children's patients selected by the feature selection model; the feature selection model uses the medical big database of children's patients stored in the historical data storage and calling unit to train and test the model in the background, and then calls out the trained model and parameters to predict the input child patient information;

[0060] The assessment of the condition of children includes the assessment of the critical illness of children and the prediction of their condition. The critical illness assessment is based on the predicted value of the important manifestations of children, combined with the pediatric critical case scoring method to score the children and give them a recommendation on whether they should be admitted to the hospital immediately, the outpatient registration department recommendation, and the name and address of the recommended hospital for admission combined with the objective conditions of the children and the external map software interface. The prediction of the condition is to obtain the consent of the family members and do not involve the privacy of the children. The predicted value of the important manifestations of the children and the collected data are integrated into an electronic medical record and sent to the recommended hospital doctor for remote diagnosis, and provide an interactive communication window between the doctor and the family members of the children.

[0061] Combine the following Figures 2 to 5 The specific process of a method for intelligent risk assessment of a child patient's condition disclosed in an embodiment of the present invention is described, which mainly includes collecting data related to the child patient's condition, analyzing the medical characteristics of the child patient, and predicting and making decisions on the child patient's condition. Figure 2 As shown in the figure, the data collection related to the condition of children patients specifically includes:

[0062] Step 110: Based on the designed data collection form for the family of the child patient, the family provides the child patient's demographic characteristics, hospital visited, illness status and simple physical sign data.

[0063] Specifically, the family members of child patients fill out the data collection form for family members of child patients through the APP or website. This form collects comprehensive, reliable and usable original data, including the demographic characteristics of child patients, hospitals visited, medical conditions and simple physical signs data.

[0064] The demographic characteristics of child patients include their personal information, family situation and location, such as name, age, address, etc. In particular, this data will be processed for privacy, and the private data will be set to an invisible form using a mask and will only be used for background model training.

[0065] The hospital visited refers to the current situation of the hospital where the child patient is being treated, including whether the child patient is currently being treated in the hospital, the grade of the hospital, the region where the patient is located, etc.

[0066] The illness condition refers to the disease currently diagnosed by the child patient, including the child patient's admission diagnosis, the child patient's admission department, whether the child patient is undergoing surgery, etc. In particular, if the child patient is not currently receiving treatment in the hospital, this data will not be filled in.

[0067] Simple child patient vital signs refer to the child patient vital signs that can be measured in ordinary home scenarios, including body temperature, heart rate, whether the limbs are cold, etc.; if the child patient is currently receiving treatment in a hospital, this data will not be filled in, because the child patient vital signs provided by the hospital are more accurate than those provided by family members.

[0068] Step 120: Based on the designed hospital data collection form, the hospital where the child patient is currently receiving treatment provides the child patient with objective in vitro tests, subjective in vitro tests and objective body fluid tests.

[0069] Specifically, the hospital where the child patient is currently receiving treatment imports images of electronic medical records or paper medical records, and the system uses regularized matching technology and text recognition technology to automatically fill out the hospital data collection form. This form collects comprehensive, reliable and available original data, including objective in vitro tests, subjective in vitro tests and objective body fluid tests of the child patient.

[0070] Among them, objective in vitro testing of children refers to testing the surface of the child's body with instruments, such as systolic blood pressure, diastolic blood pressure and respiration; subjective in vitro testing is based on the attending physician's assessment of the child's physical signs, such as cyanosis, capillary refill time and pupil light reflex; objective body fluid testing is based on blood drawing or other body fluids for testing, such as blood sugar, base excess and calcium ion concentration;

[0071] Step 130: perform data cleaning, data conversion and data filling on the collected data.

[0072] Specifically, since the collected data contain defaults and errors, and the units are inconsistent or not conducive to analysis, in order to alleviate the impact of these anomalies on the condition assessment methods of pediatric patients, the original data are converted into a form that is easy to mine.

[0073] The data cleaning mentioned above refers to merging the collected data tables and then eliminating the data records with missing data items. For example, when the child patient is not currently receiving treatment in a hospital, the data of all data items of the hospital should be eliminated; when the child patient is currently receiving treatment in a medical institution, the data of simple physical sign data items should be eliminated;

[0074] The data conversion is to convert each field in the data table according to the field characteristics, recode the lengthy discrete data, and map the time-type data to numerical values. For example, the admission diagnosis field of the child patient is "extremely low birth weight, low vitality, severe pneumonia at 28+5 weeks of gestation", which is converted to "low weight, low vitality, severe pneumonia";

[0075] The data filling is to fill in the missing data in the records, using simple mode filling for discrete data and mean filling for numerical data. For example, if the family members are uncertain about the feature of "whether the limbs of the child patient are cold", it will be left blank, resulting in the missing data of the child patient. In order to solve this situation, the mode of this feature of all child patients in the medical big data database is called as "no" to fill in the data of the child patient.

[0076] Step 140: Store the child patient data in a background database with the identity as the index and the collection time as the sequence.

[0077] Specifically, the system background sets up a medical big data database to store all child patient data. The data collected by each child patient each time will be stored after preprocessing, forming a collection time series data table with the child patient information as the main index. When the child patient data needs to be called, the data collected at all times can be called out according to the child patient information. In order to avoid duplicate names, the child patient information not only includes the child patient's name, but also the child patient's gender, date of birth and registered place of residence. This information is hashed and the obtained code is used as an index. Specifically, the storage step is to first hash the child patient information, then search whether this code exists in the main index, if it exists, then use the data collection time as the secondary index to enter the data into the database; if it does not exist, then establish this code as the main code, the data collection time as the secondary index, and enter the data into the database. The calling step is to first hash the child patient information, then search for the location where this code exists in the main index, and finally call out the data at the corresponding location.

[0078] like Figure 3 As shown in Figure 2, the medical characteristics of pediatric patients include:

[0079] Step 210: extract only the names of the physical representations related to the child patient's condition from the data collection table, and divide these representations into corresponding nodes according to medical classification.

[0080] Specifically, based on the collected medical representation data of child patients and the medical knowledge base, all medical data affecting the body of child patients are divided into four entity types, namely, cardiovascular / nervous system life representation nodes, acid / base representation nodes, biochemical detection representation nodes and blood representation nodes. Among them, the cardiovascular / nervous system life representations include systolic blood pressure, heart rate, body temperature, pupil reflex, etc., acid / base representations include acidosis, total CO2 content, pH value, PaO2, etc., biochemical detection representations include blood sugar, blood potassium, creatinine, blood urea nitrogen, etc., and blood representations include white blood cell count, prothrombin time, platelet count, etc.

[0081] Step 220: Based on the expert knowledge and the medical knowledge base, establish edges between nodes for the mutual influence of the above body representations.

[0082] Specifically, based on expert knowledge and the medical knowledge base, some of the above nodes will influence each other or represent the same condition, and edge relationships should be established between them. In addition, the edge types between different node types should be different. For example, the partial pressure of arterial carbon dioxide in the blood representation not only represents the pulmonary ventilation function, but also the respiratory factor of the body's acid-base regulation. If the measured value is greater than 50 mmHg, it indicates respiratory acidosis, and less than 35 mmHg indicates respiratory alkalosis. This has a similar effect to the meaning of base excess in the acid / base representation. They are related and influence each other. An edge connecting the blood representation and the acid / base representation should be given between these two entity nodes.

[0083] Step 230: construct a knowledge graph of risk factors for pediatric patients' conditions based on the established nodes and edges.

[0084] Specifically, the knowledge graph is composed of triplets, and a triple is composed of a head entity (head) and a tail entity (tail), as well as the association relationship (relation) between the head entity and the tail entity. The medical representation nodes of the above-mentioned child patients' conditions are used as the head entity and the tail entity, and the edges of the relationships between the above-mentioned medical representations are used as the association relationship to form a triple in the form of (head, relation, tail). Finally, all triplets are superimposed in the graph to obtain the knowledge graph of risk factors for children's conditions.

[0085] Step 240: Based on the medical big data of pediatric patients stored in the background database, the association analysis algorithm is used to calculate the degree of association between entity nodes to determine the existence of edges in the knowledge graph.

[0086] Specifically, the purpose of this step is to verify whether the relationship constructed by the medical knowledge base and expert knowledge is correct, and to use medical big data to verify the original knowledge graph. The system first calls all the child patient data stored in the historical data storage and call unit, which is called the child patient medical big data, and then calculates the Pearson correlation coefficient between the nodes in the big data as the correlation between them. According to the set threshold, if the correlation between two nodes is lower than this threshold, it means that the correlation between the two nodes is low and there should be no relationship. If there are edges between the nodes in the original knowledge graph, the corresponding edges should be deleted.

[0087] Step 250: Perform data fusion and knowledge fusion on the knowledge graph, and update the nodes and edges in the graph.

[0088] Specifically, the purpose of data fusion is to update the nodes in the knowledge graph when the data collection table in the data collection module has been updated, and the purpose of knowledge fusion is to update the edges in the knowledge graph when the medical knowledge base and expert knowledge are richer.

[0089] The data fusion described first obtains the name of the newly added table item in the data collection table, establishes a corresponding node in the graph, searches the type corresponding to this node in the medical knowledge base, and then searches whether there is a relationship between this node and all other nodes. If so, a set of triples is established, and finally the new nodes and relationships are added to the original knowledge graph based on the newly created triples.

[0090] The knowledge fusion first obtains new knowledge of a node, searches for the medical representation related to it as a node in the knowledge graph, and if so, detects the relationship between the two nodes, and establishes an edge if no edge exists; if not, abandons this piece of knowledge and continues to look for the next one.

[0091] like Figure 4 As shown, the prediction and decision-making of pediatric patients include:

[0092] Step 310: Construct a dynamic attribute graph of the child patient's condition by combining the child patient's condition representation at the current moment and the child patient's condition representation at previous moments.

[0093] Specifically, the dynamic attribute graph of the condition of children includes all the conditions of children over a period of time, not only the current moment, but also the data collected at historical moments, so it is composed of the current moment and the previous moment. The attribute graph is input into the intelligent prediction model in the form of a dynamic heterogeneous graph, including the adjacency matrix and feature matrix of the graph. Among them, the adjacency matrix represents the topological structure of the graph. Since the knowledge graph is a heterogeneous graph, that is, there are more than one node type and edge type, the adjacency matrix includes R matrices of dimension N×N, N represents the number of nodes in the graph, and R represents the number of edge types. If the value in the matrix of the corresponding edge type is 1, it means that there is an edge of this type between the corresponding two nodes in the graph. If the value is 0, it means that there is no edge of this type between the corresponding two nodes in the graph. The feature matrix represents the specific value of the node, which is obtained by normalization based on the specific value of the child's performance in medical representation.

[0094] Step 320: construct an intelligent prediction model for the child patient's condition based on the child patient's condition dynamic attribute graph, and output the predicted values ​​of the child patient's important representations.

[0095] Specifically, an intelligent prediction model for children's condition is established based on a dynamic heterogeneous graph neural network. The input of the model is the dynamic attribute graph of the child's condition, and the output is the predicted value of important characteristics of the child's condition. The parameters of the model are obtained after training using the medical big data of children stored in the background.

[0096] The dynamic heterogeneous graph neural network is implemented by heterogeneous graph neural networks, such as RGCN and GAT, plus long short-term memory networks (LSTM). The heterogeneous graph neural network is an improvement on the traditional graph neural network. Since the traditional neural network is a homogeneous graph by default, the input has only one adjacency matrix, while the heterogeneous graph has R adjacency matrices, where R represents the number of edge types. Therefore, the heterogeneous graph neural network increases the number of learning parameters, and learns R parameter matrices for each of the R adjacency matrices. The long short-term memory network is designed to learn the dynamics of data. The initial parameter matrix of the heterogeneous graph neural network of the attribute graph at a certain moment is calculated from the parameters of the network at previous moments. The specific calculation rules are specified by LSTM. The parameter matrix at the current moment is calculated from the parameter matrices at all previous moments, and the parameter matrix with a closer time distance has a greater impact on the parameter matrix at the current moment.

[0097] Among them, the important features of the child's condition are to use the feature selection model in the background to screen the representations in the child's dynamic attribute graph and select the representations that have a greater impact on the condition. The principle of the feature selection algorithm is to use the sparsity of the parameters after adding the regular term in machine learning to screen the input representations. The larger the parameter of a representation, the more important it is to the criticality of the child's condition. Conversely, it means that it is not important to the criticality of the child's condition. This model is based on the feature selection algorithm and is trained through the medical big data in the background. The criticality of the child's condition is used as the dependent variable, the representation of the child is used as the independent variable, and the parameter value is assigned to the independent variable to model the relationship between the independent variable and the dependent variable. Finally, the parameter value of the independent variable can be used to judge its importance to the dependent variable.

[0098] Step 330: assess the criticality of the child patient based on the predicted values ​​of the child patient's important characteristics, infer the type and severity of the disease, and provide medical advice.

[0099] Specifically, based on the predicted values ​​of the above-mentioned important characteristics of the child's condition, combined with a variety of pediatric critical case scoring methods, the child's current physical criticality is scored, and based on the score, two recommendations can be made for the child to seek medical treatment immediately or to choose the treatment on his own. For example, according to the score, the child's condition can be divided into five levels: very serious, serious, general, mild, and healthy. If the child's condition is general or above, the child will be advised to seek medical treatment immediately and call the external map software based on the objective conditions of the child to display the location of the recommended hospital, nearby traffic conditions, and the estimated arrival time. If it is mild or healthy, the child will be advised to choose the treatment on his own. Among them, the objective conditions of the child are provided by the child's family data collection form, including whether the child is currently receiving treatment in the hospital, the child's location, and whether there are conditions to transport the child;

[0100] In addition, after the family decides to go to the hospital, the system will, with the family's consent, prepare the above-mentioned current symptoms of the child's condition and the predicted values ​​of the important characteristics of the child's condition in the form of a structured table and send it to the outpatient doctor of the recommended hospital to achieve remote diagnosis and save time for doctors and the child's family.

[0101] A computer system disclosed in an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the steps of the method for intelligent risk assessment of a child's condition are implemented.

[0102] Those skilled in the art can understand that the technical solution of the present invention, in essence or in other words, the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer system (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in the embodiment of the present invention. The storage medium includes: various media that can store computer programs, such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

Claims

1. An intelligent risk assessment system for children's condition, characterized in that: include: The data collection module is used to collect data related to the condition from the family members and hospitals of child patients, store it after pre-processing, and support the call of historical data of child patients; The child patient condition representation data analysis module is used to extract the child patient representation related to the condition from the designed data collection table to construct entity nodes, build the relationship between entities based on expert knowledge and the medical knowledge base, and construct a knowledge graph of child patient condition risk factors; And calculate the association between entities to verify and improve the relationship between entities; The child patient condition prediction and decision-making module is used to construct a dynamic attribute graph of the child patient's condition at multiple moments based on the collected data and the knowledge graph of the child patient's condition risk factors, and then use the dynamic heterogeneous graph neural network model to learn the attribute graph, output the predicted values ​​of important representations, evaluate the degree of criticality based on the predicted values ​​of important representations, and make recommendations on the child patient's treatment strategy; The adjacency matrix of the attribute graph input into the dynamic heterogeneous graph neural network model is determined according to the topological structure of the knowledge graph provided by the child patient's condition representation data analysis module, each relationship type in the knowledge graph corresponds to an adjacency matrix, and the feature matrix is ​​determined according to the data collected at each moment by the data acquisition module; the important representation is the screened-out condition representation whose importance to the condition is greater than a set threshold.

2. The intelligent risk assessment system for children's condition according to claim 1, characterized in that: The data acquisition module comprises: The data collection unit for the family of the child patient is used to collect the disease-related data provided by the family, including the demographic characteristics of the child patient, the disease condition, the hospital visited, and the physical signs that can be measured by the family; The hospital's data collection unit is used to collect the disease-related data provided by the hospital, including the objective and subjective physical manifestations of the child patient, which are divided into objective in vitro testing, subjective in vitro testing and objective body fluid testing data; the objective in vitro testing is to use instruments to test the body surface of the child patient, the subjective in vitro testing is based on the attending physician's assessment of the child patient's physical signs, and the objective body fluid testing is based on blood drawing or other body fluids for testing; A historical data storage and calling unit is used to establish an index in the database based on the child patient identity information, and store the child patient information at a location corresponding to the index according to the time of collection, so as to form a data table of the child patient data time series with the child patient identity information as the index; The data preprocessing unit is used to clean, convert and fill the collected data.

3. The intelligent risk assessment system for children's condition according to claim 1, characterized in that: The child patient condition characterization data analysis module comprises: The disease risk factor knowledge graph construction unit is used to extract entity nodes from the child patient family data collection table and the hospital data collection table and establish edge relationships between the nodes to construct a knowledge graph; the extracted entity nodes are the names of physical representations that affect the child patient's disease or may bring risks to the child patient's physical health from the collection table, and the edge relationships of the nodes are established based on expert knowledge and medical knowledge bases to model the mutual influence of the extracted physical representations, and connect the nodes that influence each other or represent the same disease condition; The knowledge graph deep analysis unit is used to calculate the degree of association between entity nodes based on the medical big data of pediatric patients stored in the database using the association analysis algorithm, determine the existence of edges in the knowledge graph, and delete edges with a degree of association lower than the set threshold; The knowledge graph deep fusion unit is used to update the knowledge graph, including data fusion and knowledge fusion; the data fusion is based on the updated child patient family data collection form and hospital data collection form, and updates the number and / or type of nodes in the child patient condition risk factor knowledge graph; the knowledge fusion is based on the updated expert knowledge and medical knowledge base, and updates the number and / or type of edges in the child patient condition risk factor knowledge graph.

4. The intelligent risk assessment system for children's condition according to claim 1, characterized in that: The child patient condition prediction and decision-making module includes: A child patient's condition dynamic attribute graph construction unit is used to construct a child patient's condition dynamic attribute graph, including the child patient's condition representation at the current moment and the child patient's condition representation at previous moments; wherein the child patient's condition representation attribute graph at the current moment is established based on the child patient's condition risk factor knowledge graph, wherein the attribute value includes the specific value or severity description of the corresponding representation; the attribute value in the child patient's condition representation attribute graph at previous moments is provided by the child patient's historical data, and is stacked and merged with the child patient's condition representation attribute graph at the current moment in a time series to form a dynamic graph; A child patient condition prediction model unit, used for inputting a child patient condition dynamic attribute graph into a child patient condition intelligent diagnosis model to predict the value of an important representation of the child patient, wherein the child patient condition intelligent diagnosis model is a dynamic heterogeneous graph neural network model trained using historical data provided by a data acquisition module; The child patient condition assessment unit is used to score the criticality of child patients based on the predicted values ​​of the child patients' important characteristics and the pediatric critical patient scoring table, and give immediate medical treatment or self-selected medical treatment recommendations based on the scores; if immediate medical treatment is recommended, the map interface will be connected to find nearby hospitals and route traffic conditions and recommend hospitals based on the child patients' objective conditions.

5. The intelligent risk assessment system for children's condition according to claim 4, characterized in that: The child patient condition assessment unit is also used to send the predicted values ​​of the child patient's important manifestations and the collected data to the doctor of the hospital selected by the family for remote diagnosis, with the consent of the family and without violating the privacy of the child patient.

6. The intelligent risk assessment system for children's condition according to claim 1, characterized in that: The important representations are screened out according to a feature selection model based on machine learning. The feature selection model is trained through historical data of child patients, takes the criticality of child patients as the dependent variable, and the representations of child patients as the independent variable, models the relationship between the independent variable and the dependent variable, assigns parameter values ​​to the independent variables, and finally measures the importance of the independent variables with the size of the parameter values.

7. An intelligent risk assessment method for a pediatric patient's condition, characterized in that: The steps include: Collecting data related to the condition from the family members of the child patients and the hospital, pre-processing and storing the data, the stored data including the historical data of the child patients collected at different times; Extract the representation of children's patients related to their conditions from the designed data collection table to construct entity nodes, build the relationship between entities based on expert knowledge and the medical knowledge base, and construct a knowledge graph of risk factors for children's conditions; And calculate the association between entities to verify and improve the relationship between entities; Based on the collected data and the knowledge graph of risk factors of children's conditions, a dynamic attribute graph of children's conditions at multiple times is constructed. Then, the attribute graph is learned using a dynamic heterogeneous graph neural network model, and the predicted values ​​of important representations are output. Based on the predicted values ​​of important representations, the degree of criticality is assessed, and treatment strategies for children are recommended. The adjacency matrix of the attribute graph input into the dynamic heterogeneous graph neural network model is determined according to the topological structure of the knowledge graph provided by the child patient's condition representation data analysis module, each relationship type in the knowledge graph corresponds to an adjacency matrix, and the feature matrix is ​​determined according to the data collected at each moment by the data acquisition module; the important representation is the screened-out condition representation whose importance to the condition is greater than a set threshold.

8. The intelligent risk assessment method for a child's condition according to claim 7, characterized in that: Construction and update of the knowledge graph of risk factors for pediatric patients, including: Entity nodes are extracted from the child patient family data collection table and the hospital data collection table, and edge relationships between the nodes are established to construct a knowledge graph; the extracted entity nodes are extracted from the collection table, and the names of physical representations that affect the child patient's condition or may bring risks to the child patient's physical health are extracted; the edge relationships of the nodes are established based on expert knowledge and a medical knowledge base, and the mutual influence of the extracted physical representations is modeled, and nodes that influence each other or represent the same condition are connected; Based on the medical big data of pediatric patients stored in the database, the association analysis algorithm is used to calculate the degree of association between entity nodes, determine the existence of edges in the knowledge graph, and delete edges with a degree of association lower than the set threshold; Based on the updated child patient family data collection form and hospital data collection form, update the number and / or type of nodes in the child patient condition risk factor knowledge graph; based on the updated expert knowledge and medical knowledge base, update the number and / or type of edges in the child patient condition risk factor knowledge graph.

9. The intelligent risk assessment method for a child's condition according to claim 7, characterized in that: The important representations predicted by the dynamic heterogeneous graph neural network model are screened out according to the feature selection model based on machine learning. The feature selection model is trained by the historical data of child patients, takes the criticality of child patients as the dependent variable, and the representations of child patients as the independent variable, models the relationship between the independent variable and the dependent variable, assigns parameter values ​​to the independent variables, and finally measures the importance of the independent variables by the size of the parameter values.

10. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is loaded into the processor, the intelligent risk assessment method for the condition of a child patient according to any one of claims 7 to 9 is implemented.

Images