Personal electronic medical record-based uncertainty knowledge graph automatic construction method

Abstract

The invention discloses a method for automatically constructing an uncertain knowledge graph based on a personal electronic medical record. According to the method, firstly, a comprehensive uncertainty mapping knowledge domain ontology model is defined, the model comprises entities of diseases, symptoms, examinations and the like and relationships thereof, and confidence attributes are introduced to quantify the possibility of the relationships; thirdly, designing an efficient mapping mechanism, and systematically mapping the uncertain knowledge in the individual medical record knowledge graph into the uncertain knowledge graph; in addition, in order to improve processing efficiency, the invention further provides an optimization algorithm, parallel processing is performed by analyzing node degree sorting, calculation overhead is reduced, repeated calculation is avoided, and therefore the construction speed of the knowledge graph is increased. According to the method, the capacity of capturing complex relations is enhanced, the decision support level is improved, standardized management and informatization development of medical data are promoted, and a powerful tool is provided for achieving precise medical treatment.

Description

Technical Field

[0001] Based on the fields of knowledge graphs and medical information processing technology, the present invention studies a method for automatically constructing a medical knowledge graph containing uncertainty based on personal electronic medical records (EMR). This method aims to analyze and extract uncertainty information in individual electronic medical records, and proposes an efficient mapping mechanism to extract uncertainty knowledge in individual knowledge graphs into uncertainty knowledge graphs. Background Art

[0002] The patient diagnosis and treatment information recorded in the electronic medical record (EMR) is an important record of the doctor's diagnosis and treatment process. It contains the relationship between multiple elements such as symptoms, diseases, causes, and treatments, and is important clinical diagnosis and treatment knowledge. These historically formed clinical diagnosis and treatment knowledge will provide important support for assisting doctors in disease diagnosis, predicting disease progression, optimizing treatment plans, and carrying out personalized medicine and precision medicine. Clinical diagnosis and treatment knowledge is generated based on each personal medical record information, and its characteristics are certain empirical and uncertain. Using uncertainty knowledge graphs to express it will be beneficial to subsequent knowledge reasoning and application. Therefore, effectively generating uncertainty knowledge graphs that express clinical diagnosis and treatment knowledge is the key to exerting its application value. At present, some studies have developed methods for constructing uncertainty knowledge graphs to a certain extent, and have also been meaningfully applied in some fields, but no relevant research has been found on the automatic construction of uncertainty knowledge graphs based on existing deterministic knowledge graphs. Therefore, the present invention attempts to design an uncertainty knowledge graph ontology definition that expresses clinical diagnosis and treatment knowledge and a mapping method for generating uncertainty knowledge graphs from personal medical record knowledge graphs. Summary of the invention

[0003] The main purpose of the present invention is to propose a mapping method that can generate an uncertainty knowledge graph from a personal medical record knowledge graph, so as to provide a method for constructing an uncertainty knowledge graph, which is convenient for extracting and utilizing the clinical diagnosis and treatment knowledge contained in electronic medical records.

[0004] To achieve the above objectives, the adopted technical solution mainly includes three parts:

[0005] Part I: Uncertainty Knowledge Graph Ontology Definition. This paper first analyzes the sources and characteristics of uncertainty information in electronic medical records and proposes an uncertainty knowledge graph ontology model. The model includes diagnosis and treatment related entities (such as diseases, symptoms, examinations, drugs, etc.) and the relationships between them, and introduces a confidence attribute for each relationship to indicate the possibility of association between entities.

[0006] Part II: Proposal of mapping mechanism. This paper proposes an efficient mapping mechanism to map the uncertainty knowledge in the individual medical record knowledge graph to the uncertainty knowledge graph. By analyzing the structural differences between the individual knowledge graph and the uncertainty knowledge graph, the mapping rules of nodes and relationships are designed, and an algorithm based on statistics and reasoning is proposed to generate the confidence of the relationship.

[0007] Part III: Design of mapping algorithm. This paper also designs an optimized mapping algorithm, which improves the mapping efficiency by reducing computational overhead and avoiding repeated computation. The algorithm determines the mapping order according to the degree of the graph nodes and significantly accelerates the construction process of the knowledge graph through parallel computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is an individual medical record knowledge graph ontology model.

[0009] Figure 2 This is an example of a patient individual knowledge graph.

[0010] Figure 3 It is an uncertainty knowledge graph ontology model. DETAILED DESCRIPTION

[0011] The following is a detailed description of the mapping method of the present invention for generating an uncertainty knowledge graph from a personal medical record knowledge graph in conjunction with the accompanying drawings:

[0012] Step 1: Obtain the original data of the individual electronic medical record knowledge graph.

[0013] Step 1.1: Identify data sources, including but not limited to hospital information systems (HIS), laboratory information management systems (LIMS), etc., which store patients' electronic medical record data.

[0014] Step 1.2: Clean the collected data to remove erroneous data, duplicate records, and irrelevant information to ensure data quality.

[0015] Step 1.3: According to Figure 1 The individual medical record knowledge graph ontology model shown in the figure stores the data in the neo4j graph database to build an individual medical record knowledge graph. For example, Figure 2 shown.

[0016] Step 2: Design an uncertainty knowledge graph ontology model based on individual electronic medical records.

[0017] Step 2.1: Design entity types for the uncertainty knowledge graph ontology. In the individual knowledge graph ontology model, entities and relationship attributes with group properties can become uncertainty knowledge graph entity types. During the design, entities with group properties will become new entity types of the uncertainty knowledge graph, where the attributes related to a specific patient's information will be removed, and the remaining attributes will be used as attributes of the new entity; for relationship attributes with group properties, they will be independently used as new entity types, and description fields will be added as entity attributes.

[0018] Step 2.2: Design relationship types for the uncertainty knowledge graph. According to all entity types of the uncertainty knowledge graph obtained in the above steps, obtain the path between every two types of entities in the original individual knowledge graph. If the entity type is obtained by the relationship attribute, the entity farthest from the other end of the path is taken as the path node. Finally, the relationship between entity types with a path length of not less than 2 will be used as a candidate relationship type to determine the relationship type in the uncertainty knowledge graph ontology.

[0019] Step 2.3: Organize the entities, relationships, and attributes obtained above into triples to form an uncertainty knowledge graph ontology data model, such as Figure 3 shown.

[0020] Step 3: Construct a mapping scheme from the personal medical record knowledge graph to the uncertainty knowledge graph.

[0021] Step 3.1: Design a mapping scheme for the uncertainty knowledge graph nodes. The nodes in the uncertainty knowledge graph can be obtained from the personal medical record knowledge graph in two ways:

[0022] (1) Direct node mapping, that is, directly mapping nodes in the personal medical record knowledge graph to nodes in the uncertainty knowledge graph. First, determine which nodes can be directly mapped. For example, nodes such as symptoms, diseases, examinations, and tests and their attributes can be directly mapped to corresponding nodes in the uncertainty knowledge graph. For each directly mapped node, only its basic attributes in the individual knowledge graph are retained, and information related to a specific patient is removed. For example, for the "diabetes" node, we may only retain its name and ID, and remove the diagnosis date or treatment record of a specific patient;

[0023] (2) Relationship attribute mapping, that is, the nodes of the uncertainty knowledge graph come from the attributes of the relationships in the personal medical record knowledge graph. For example, the nodes of the examination results and test results of the uncertainty knowledge graph come from Figure 1The Result attribute of the relationship between the examinations and tests performed in the patient's visit is used to express the examination results or test indicators of the patient's visit. These relationship attributes are converted into new entity types, and necessary description fields are added to them to enhance their expressiveness. This helps to better capture the uncertainty and complexity in electronic medical records.

[0024] Step 3.2: Design a mapping scheme for the uncertainty knowledge graph relationship. The relationship between two nodes in the uncertainty knowledge graph is mainly generated by the relationship between indirect related nodes in the personal medical record knowledge graph, which can be divided into two cases: one is that the two indirect related nodes in the personal medical record knowledge graph directly correspond to the adjacent nodes with direct relationship in the uncertainty knowledge graph. For example, the direct relationship between symptoms and diseases in the uncertainty knowledge graph is generated by Figure 1 The symptom node in the personal medical history knowledge graph is generated through the relationship between the medical experience node and the disease node. Another case is that one of the two adjacent nodes in the uncertainty knowledge graph is mapped by the attribute of the relationship in the personal medical history knowledge graph, such as the test result node in the uncertainty knowledge graph corresponds to the attribute "test result" of the relationship between the medical experience node and the test node in the personal knowledge graph.

[0025] Step 4: The process of constructing an uncertainty knowledge graph based on the personal medical record knowledge graph.

[0026] Step 4.1: Degree calculation. Calculate the degree of all nodes in the uncertainty knowledge graph ontology structure designed in step 2, that is, the number of connections between each node and other nodes.

[0027] Step 4.2: Node sorting. Sort the nodes based on their degrees to form an ordered set L. If there are nodes with the same degree, they are further sorted according to the sum of the degrees of other nodes they are directly associated with. This sorting method helps to prioritize nodes with more associations in subsequent processing, thereby reducing unnecessary repeated visits and speeding up the entire construction process.

[0028] Step 4.3: For each node element in L, obtain other nodes associated with it and form a set of triples. According to the order of the nodes after the above sorting, select a node G from the set L in turn. max Then, find the max The set of all directly related nodes G rel , for each G max and its related node G rel For each pair of nodes in G, the relationship between them is identified according to the ontology model of the individual electronic medical record knowledge graph, and a series of triples (entity-relationship-entity) are formed. max represents the disease node, and G relIf it contains nodes such as symptoms and examinations, you need to identify all possible relationships between these nodes and record them.

[0029] Step 4.4: Map each triple according to the mapping scheme defined in step 3. Specifically, extract the corresponding instance data from the individual medical record knowledge graph and convert it into the uncertainty knowledge graph format. In this process, for nodes that can be directly mapped (such as diseases, symptoms, etc.), only retain their basic attributes in the individual knowledge graph, and remove the part involving specific patient information; for new node types derived from relationship attributes (such as examination results, test results), they need to be independently treated as new entity types, and add description fields as entity attributes; calculate and assign confidence values ​​(PosConfidence and NegConfidence) for each pair of node relationships based on statistical methods. Finally, store the mapped data in the Neo4j database. Ensure that each newly generated triple correctly reflects the information in the original individual medical record knowledge graph and conforms to the ontology definition of the uncertainty knowledge graph.

[0030] The present invention has the following beneficial effects:

[0031] The present invention proposes a method for automatically constructing a medical knowledge graph containing uncertainty based on personal electronic medical records (EMRs), by analyzing and extracting uncertainty information in individual electronic medical records and mapping it into a structured uncertainty knowledge graph. The method first defines an uncertainty knowledge graph ontology model that comprehensively covers multiple relationships such as diseases, symptoms, and examinations, and introduces confidence attributes to quantify the possibility of association between entities. Then, an efficient mapping mechanism is designed to extract and map the uncertainty knowledge in the individual medical record knowledge graph into the uncertainty knowledge graph to ensure the accuracy and completeness of data conversion. In addition, the optimized mapping algorithm significantly improves the processing efficiency by reducing computational overhead and avoiding repeated calculations, and uses parallel computing to accelerate the construction process of the knowledge graph. This method not only improves the accuracy and personalization level of clinical diagnosis and treatment, enhances the ability to capture complex relationships and support decision-making, but also promotes the standardized management of medical data, and promotes the development of interdisciplinary cooperation and medical informatization. Finally, the present invention provides doctors with more scientific and reasonable diagnosis and treatment suggestions, helping to achieve precision medicine and personalized health management.

Claims

1. A method for automatically constructing an uncertainty knowledge graph based on personal electronic medical records, characterized in that: The following steps are involved: Step 1: Obtain the original data of the individual electronic medical record knowledge graph; Step 1.1: Identify data sources, including hospital information systems (HIS) and laboratory information management systems (LIMS), which store patients’ electronic medical records. Step 1.2: Clean the collected data to remove erroneous data, duplicate records, and irrelevant information to ensure data quality; Step 1.3: According to the individual medical record knowledge graph ontology model, store the data in the neo4j graph database to build the individual medical record knowledge graph; Step 2: Design an uncertainty knowledge graph ontology model based on individual electronic medical records; Step 2.1: Design entity types for the uncertainty knowledge graph ontology. In the individual knowledge graph ontology model, entities and relationship attributes with group properties can become uncertainty knowledge graph entity types. When designing, entities with group properties will become new entity types of the uncertainty knowledge graph, in which the attributes related to a specific patient's information will be removed, and the remaining attributes will be used as attributes of the new entity. For relationship attributes with group properties, they will be independently used as new entity types, and description fields will be added as entity attributes. Step 2.2: Design relationship types for the uncertainty knowledge graph. According to all entity types of the uncertainty knowledge graph obtained in the above steps, obtain the path between each two types of entities in the original individual knowledge graph. If the entity type is obtained by the relationship attribute, the entity farthest from the other end of the path is taken as the path node. Finally, the relationship between entity types with a path length of not less than 2 will be taken as a candidate relationship type. Step 2.3: Organize the entities, relationships and attributes obtained above into triples to form an uncertainty knowledge graph ontology data model; Step 3: Construct a mapping scheme from the personal medical record knowledge graph to the uncertainty knowledge graph; Step 3.1: Design a mapping scheme for the uncertainty knowledge graph nodes. The nodes in the uncertainty knowledge graph can be obtained from the personal medical record knowledge graph in two ways: (1) Direct node mapping, that is, directly mapping nodes in the personal medical record knowledge graph to nodes in the uncertainty knowledge graph. First, determine which nodes to map directly; directly map symptom, disease, examination, and test nodes and their attributes to corresponding nodes in the uncertainty knowledge graph. For each directly mapped node, only its basic attributes in the individual knowledge graph are retained, and information related to a specific patient is removed; (2) Relationship attribute mapping, that is, the nodes of the uncertainty knowledge graph come from the attributes of the relationships in the personal medical record knowledge graph; these relationship attributes are converted into new entity types and description fields are added to them; Step 3.2: Design a mapping scheme for the uncertainty knowledge graph relationship. The relationship between two nodes in the uncertainty knowledge graph is mainly generated by the relationship between indirect related nodes in the personal medical record knowledge graph, which can be divided into two cases: one is that two indirect related nodes in the personal medical record knowledge graph directly correspond to adjacent nodes with direct relationship in the uncertainty knowledge graph; the other is that one of the two adjacent nodes in the uncertainty knowledge graph is mapped by the attribute of the relationship in the personal medical record knowledge graph, such as the test result node in the uncertainty knowledge graph corresponds to the attribute "test result" of the relationship between the medical experience node in the personal knowledge graph and the test node; Step 4: The process of constructing an uncertainty knowledge graph from the personal medical record knowledge graph; Step 4.1: Degree calculation: Calculate the degree of all nodes in the uncertainty knowledge graph ontology structure designed in step 2, that is, the number of connections between each node and other nodes; Step 4.2: Node sorting: Sort nodes based on their degrees to form an ordered set L; if there are nodes with the same degree, further sort them based on the sum of the degrees of other nodes they are directly associated with; Step 4.3: For each node element in L, obtain other nodes associated with it and form a set of triples; select a node Gmax from the set L in turn according to the order of the nodes after the above sorting; then, find all the node sets Grel directly related to Gmax, and for each pair of nodes in Gmax and its related node Grel, identify the relationship between them according to the ontology model of the individual electronic medical record knowledge graph, and form a series of triples, namely entity-relationship-entity; Step 4.4: According to the mapping scheme defined in step 3, map each triple; extract the corresponding instance data from the individual medical record knowledge graph and convert it into the uncertainty knowledge graph format; in this process, for directly mapped nodes including diseases and symptoms, only retain their basic attributes in the individual knowledge graph and remove the part involving specific patient information; for new node types derived from relationship attributes including examination results and test results, they need to be independently treated as new entity types, and add description fields as entity attributes; calculate and assign confidence values ​​for the relationship between each pair of nodes based on statistical methods; finally, store the mapped data in the Neo4j database.

Images