Medical knowledge graph construction and reasoning method and system, medium and equipment

Abstract

The invention provides a medical knowledge graph construction and reasoning method and system, a medium and equipment. The medical knowledge graph construction and reasoning method comprises the steps that to-be-queried content and a constructed medical knowledge graph are acquired; according to the to-be-queried content, the constructed medical knowledge graph is subjected to recognition reasoning through the graph neural network, and a query result is determined and comprises the entity relation of the to-be-queried content in the medical knowledge graph. According to the method, the knowledge graph is reasoned by adopting the graph convolutional network, the graph data structure can be effectively processed, the complex relationship between entities can be captured, the problem of incomplete information can be solved, and the expression and reasoning efficiency of the medical knowledge graph is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of medical knowledge graph construction and reasoning technology, and specifically, to a medical knowledge graph construction and reasoning method, system, medium and equipment. Background Art

[0002] Most existing medical knowledge management systems use traditional database technology for management and storage, which is inadequate when faced with complex medical relationships. For example, when it is necessary to express the complex relationships between diseases, symptoms, and treatment methods, traditional databases are difficult to intuitively represent these relationships and cannot effectively support in-depth analysis of these relationships.

[0003] In addition, because medical knowledge often involves multi-level and multi-dimensional information, such as the mutual influence of different diseases, the interaction between drugs and symptoms, etc., traditional database systems often find it difficult to provide a medical knowledge graph to represent these complex and interrelated entities. This makes it difficult to fully utilize these potential knowledge associations when making decision support in disease diagnosis, treatment plan formulation, etc., thus affecting the quality and efficiency of medical services.

[0004] On the other hand, traditional databases have the following problems: when processing large-scale data sets, the reasoning efficiency is low, and it is difficult to fully utilize the potential knowledge associations between data. Summary of the invention

[0005] In view of the defects in the prior art, the purpose of the present invention is to provide a medical knowledge graph construction and reasoning method, system, medium and equipment.

[0006] The medical knowledge graph construction and reasoning method provided by the present invention includes: obtaining content to be queried and a constructed medical knowledge graph; based on the content to be queried, identifying and reasoning the constructed medical knowledge graph through a graph neural network to determine the query result, and the query result includes: the entity relationship of the content to be queried in the medical knowledge graph.

[0007] It is not difficult to understand that the present invention uses graph neural networks to perform recognition and reasoning on the constructed medical knowledge graph, which can not only effectively process graph data structures, but also capture complex relationships between entities and handle problems with incomplete information, significantly improving the representation and reasoning efficiency of medical knowledge graphs; further, the present invention can integrate information from multiple data sources through knowledge graphs, such as genomics, pathology reports, clinical trial data, etc., to help doctors diagnose diseases more accurately, especially for the identification of rare diseases or complex diseases.

[0008] Furthermore, the process of constructing a medical knowledge graph includes: acquiring medical research literature data; mapping the medical research literature data into an embedding vector through an embedding model, and then performing bidirectional semantic feature learning on the embedding vector through an encoder to obtain a dynamic word vector; parsing the dynamic word vector representing the ambiguity of a word through a decoder to capture long-distance dependencies, temporal correlations and global features between data elements, extracting entities based on the long-distance dependencies, temporal correlations and global features between data elements, and determining entity relationship triplets, thereby realizing the construction of a medical knowledge graph.

[0009] It is not difficult to understand that by using embedding models, encoders and decoders, a large amount of complex medical literature data can be effectively processed, useful information can be extracted and organized into a structured knowledge graph, which helps to quickly integrate scattered knowledge points and form a systematic knowledge system. Due to the use of two-way semantic feature learning, the model can better understand the meaning of words in different contexts, improving the accuracy and reliability of subsequent knowledge reasoning.

[0010] Furthermore, the embedding model is a model trained by a partial word masking strategy, and the training process includes: performing word segmentation processing on the medical research literature data to obtain multiple word fragments; randomly selecting a preset proportion of word fragments from the multiple word fragments for masking to obtain masked word fragments and unmasked word fragments; mapping the unmasked word fragments into embedding vectors, and calculating the context information of the embedding vectors through a self-attention mechanism; predicting the masked word fragments based on the context information to obtain prediction results; and adjusting the parameters of the embedding model based on the prediction results until the prediction accuracy of the embedding model meets the preset requirements.

[0011] Furthermore, the constructed medical knowledge graph is stored in a graph database, and the medical knowledge graph is queried through the graph database.

[0012] It is not difficult to understand that graph databases are used to process complex relational queries and can perform query operations involving multiple entities and multiple relations with higher efficiency than traditional relational databases. Furthermore, by utilizing the rich information in graph databases, personalized treatment plans can be customized for each patient's specific situation to improve treatment effectiveness.

[0013] Furthermore, performing recognition and reasoning on the constructed medical knowledge graph through the graph neural network includes: mapping the entities and relationships in the knowledge graph into a graph data structure through the graph neural network, and determining the entity relationships in the constructed medical knowledge graph, wherein the nodes in the graph data structure represent entities and the edges represent the relationships between entities.

[0014] It is not difficult to understand that graph neural networks can capture the complex relationships between nodes in graph data structures, which allows the model to learn multi-level dependencies between entities without explicit rules, thereby better understanding the information in the medical knowledge graph. Through graph data structure modeling, graph neural networks can use the information of neighboring nodes to more accurately represent entities, and can learn the potential transmission mode or etiology mechanism of the disease from existing data, helping doctors to make more accurate diagnoses and develop more personalized treatment plans based on individual differences.

[0015] Furthermore, the graph neural network includes a graph convolutional network, which performs feature learning on the graph data structure and infers the entity relationship in the medical knowledge graph, and the expression is:

[0016]

[0017] Among them, H (l) H represents the node feature matrix of the lth layer, indicating the attributes of the entity corresponding to the node or its relationship with other entities; (l+1) Represents the node feature matrix of the l+1th layer, indicating the update of node features after a graph convolution operation; Represents the adjacency matrix A of the graph plus the identity matrix I N , used to represent the direct relationship between entities in the medical knowledge graph; for The diagonal matrix of the degree matrix is ​​used for normalization to ensure the stability of the training process; W (l) represents the weight matrix of the lth layer, which is learned through the training process and used to update the node features; σ is the activation function, which is used to introduce nonlinearity, enhance the expressiveness of the model, represent the node update process or the complex relationship between nodes, and thus improve the accuracy of the reasoning results.

[0018] Furthermore, it also includes: displaying the query results and graph exploration paths on the user interface.

[0019] It is not difficult to understand that by displaying query results and exploration paths in a graphical way, users can see the relationship between entities at a glance, making complex medical information easier to understand and digest. Furthermore, doctors and other medical professionals can confirm the basis for diagnosis or understand the logical chain behind a treatment plan by viewing the relationship chain in the graph, so as to make more reasonable clinical decisions.

[0020] The medical knowledge graph construction and reasoning system provided by the present invention includes:

[0021] Knowledge graph construction module: obtain the content to be queried and the constructed medical knowledge graph;

[0022] Reasoning and query module: Based on the content to be queried, the constructed medical knowledge graph is identified and inferred through the graph neural network to determine the query results. The query results include: the entity relationship of the content to be queried in the medical knowledge graph.

[0023] According to the computer-readable storage medium storing a computer program provided by the present invention, when the computer program is executed by a processor, the steps of the medical knowledge graph construction and reasoning method are implemented.

[0024] The electronic device provided according to 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 executed by the processor, the steps of the medical knowledge graph construction and reasoning method are implemented.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention uses graph neural networks to perform recognition and reasoning on the constructed medical knowledge graphs, which can not only effectively process graph data structures, but also capture complex relationships between entities and handle problems with incomplete information, significantly improving the representation and reasoning efficiency of medical knowledge graphs. The present invention can integrate information from multiple data sources through knowledge graphs, such as genomics, pathology reports, clinical trial data, etc., to help doctors diagnose diseases more accurately, especially for the identification of rare diseases or complex diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0028] Figure 1 This is a flow chart of the medical knowledge graph construction and reasoning method of the present invention;

[0029] Figure 2 Constructing knowledge graph fluid for the present invention;

[0030] Figure 3 This is a block diagram of the medical knowledge graph construction and reasoning system of the present invention. DETAILED DESCRIPTION

[0031] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0032] The present invention provides a medical knowledge graph construction and reasoning method, including: obtaining content to be queried and a constructed medical knowledge graph; identifying and reasoning the constructed medical knowledge graph through a graph neural network to determine the entity relationship in the constructed medical knowledge graph; and obtaining a query result of the medical knowledge graph based on the content to be queried.

[0033] It is not difficult to understand that the present invention uses graph neural networks to perform recognition and reasoning on the constructed medical knowledge graph, which can not only effectively process graph data structures, but also capture complex relationships between entities and handle problems with incomplete information, significantly improving the representation and reasoning efficiency of medical knowledge graphs. Furthermore, the present invention can integrate information from multiple data sources through knowledge graphs, such as genomics, pathology reports, clinical trial data, etc., to help doctors diagnose diseases more accurately, especially for the identification of rare diseases or complex diseases.

[0034] like Figure 1 , the present invention proposes a medical knowledge graph construction and reasoning method, comprising the following steps:

[0035] Step S1: Obtain the content to be queried and the constructed medical knowledge graph.

[0036] In the embodiments of the present application, the content to be queried may be information such as the symptoms, causes, and treatment methods of a certain disease, the mechanism of action, indications, and side effects of a specific drug, or the process and risks of a medical procedure or surgery. The constructed medical knowledge graph records the medical entities and the relationships between them based on the existing medical professional dictionary.

[0037] Step S2: Based on the content to be queried, the constructed medical knowledge graph is identified and inferred through the graph neural network to determine the query result, which includes: the entity relationship of the content to be queried in the medical knowledge graph.

[0038] Specifically, the entities and relationships in the knowledge graph are mapped into a graph data structure through a graph neural network to determine the entity relationships in the constructed medical knowledge graph, where the nodes in the graph data structure represent entities and the edges represent the relationships between entities.

[0039] The graph neural network includes a graph convolutional network, which maps the standardized entities and relationships in the knowledge graph into a graph data structure, in which nodes represent entities and edges represent the relationships between entities. By training the graph convolutional network, the node representation is optimized to capture the relationship between entities and determine the entity relationship of the constructed knowledge graph. A key feature of the graph convolutional network is that it can perform effective feature learning on the graph data structure and infer the entity relationship in the medical knowledge graph, which is expressed as:

[0040]

[0041] Among them, H (l) H represents the node feature matrix of the lth layer, indicating the attributes of the entity corresponding to the node or its relationship with other entities; (l+1) Represents the node feature matrix of the l+1th layer, indicating the update of node features after a graph convolution operation; Represents the adjacency matrix A of the graph plus the identity matrix I N (representing self-join), which is used to represent the direct relationship between entities in the medical knowledge graph; for The diagonal matrix of the degree matrix is ​​used for normalization to ensure the stability of the training process; W (l) represents the weight matrix of the lth layer, which is learned through the training process and is used to update the node features; σ represents the activation function (such as ReLU), which is used to introduce nonlinearity, enhance the expressiveness of the model, represent the node update process or the complex relationship between nodes, and thus improve the accuracy of the inference results.

[0042] In an embodiment of the present application, the graph neural network uses known entities and relationships in the knowledge graph as input during the training phase, and adjusts the model parameters through the back propagation algorithm so that the relationship between entities can be accurately predicted; after the training is completed, the graph neural network can be used to identify missing or unclear relationships in the knowledge graph. In the application phase of the graph neural network, the missing or unclear relationships in the identified knowledge graph need to be verified to ensure that they are logically reasonable and consistent with the actual knowledge; once the missing or unclear relationships in the identified knowledge graph are verified to be valid, they will be added to the graph database to enrich and improve the existing knowledge graph. The whole process can be an iterative process, that is, by continuously applying the graph neural network for prediction and updating, the quality and coverage of the knowledge graph are gradually improved. In this way, the graph neural network can not only help identify and supplement the missing relationships in the knowledge graph, but also enhance its reasoning ability, making the knowledge graph a more powerful and flexible data structure that can be used in a variety of application scenarios, such as recommendation systems, question-answering systems, etc.

[0043] Furthermore, the constructed medical knowledge graph and the determined entity relationships are stored in a graph database, and the medical knowledge graph is queried through the graph database.

[0044] Step S3: Display the query results and graph exploration path on the user interface.

[0045] Specifically, in some embodiments, displaying the query results and the graph exploration path on the user interface includes: providing an intuitive graphical interface to visualize the query results. For example, entities in the query results are represented graphically, and the relationship paths between the entities are dynamically displayed according to the exploration process. In this way, not only can the query results be clearly displayed to the user, but also the user can be helped to better understand and analyze the complex relationships between the data.

[0046] like Figure 2 In this embodiment, the process of constructing the medical knowledge graph includes:

[0047] Step S1.1: Obtain medical research literature data.

[0048] Medical research literature data can be obtained from academic databases and journals, publications of professional societies and institutions, research reports released by governments and international organizations, academic repositories of universities and research institutions, social media and professional network platforms, library services, and clinical trial registries.

[0049] Step S1.2: Map the medical research literature data into an embedding vector through an embedding model, and then perform bidirectional semantic feature learning on the embedding vector through an encoder to obtain a dynamic word vector.

[0050] Step S1.3: The dynamic word vector representing the ambiguity of the word is parsed through the decoder to capture the long-distance dependencies, temporal correlations and global features between data elements. Entities are extracted and entity relationship triplets are determined based on the long-distance dependencies, temporal correlations and global features between data elements, thereby realizing the construction of a medical knowledge graph.

[0051] Specifically, the embedding model is a model trained by a partial word masking strategy, and the training process includes: performing word segmentation processing on medical research literature data to obtain multiple word fragments; randomly selecting a preset proportion of word fragments from the multiple word fragments for masking to obtain masked word fragments and unmasked word fragments; mapping the unmasked word fragments into embedding vectors, and calculating the context information of the embedding vectors through a self-attention mechanism; predicting the masked word fragments based on the context information to obtain prediction results; and adjusting the parameters of the embedding model based on the prediction results until the prediction accuracy of the embedding model meets the preset requirements.

[0052] The embedding model is an architecture with a built-in 12-layer bidirectional decoder. Each unit consists of a feedforward neural network and a multi-head self-attention mechanism. Its self-attention mechanism can focus on information at different positions when processing the input sequence, enhance global understanding, reduce overfitting at specific positions, and improve the generalization ability of the model. The calculation method of its attention is as follows:

[0053]

[0054] Among them, Q represents the inference matrix, K represents the key matrix, V represents the value matrix, and d k is the dimension of K matrix; T is the matrix transpose symbol;

[0055] According to the calculation formula, the self-attention mechanism is parsed for the vectorized input. The correlation between elements is measured by calculating the attention weight distribution and the corresponding scaling control and weight calculation are performed to obtain the importance information of each word to the entire text, thereby learning the semantic features of the context at a deeper level.

[0056] Specifically, the process of entity extraction and determination of entity relationship triples is to use the decoder's self-attention mechanism to capture the long-distance dependencies between words in the text. At the same time, through sequence modeling and time embedding technology, we can further enhance the model's ability to understand time series data, which is particularly important for processing medical events with time attributes. Next, after capturing these high-level abstract features, the named entity recognition (NER) algorithm is applied to identify entities in the text, and combined with dependency syntax analysis and other methods, relational triples are extracted from the relationship between entities, that is, the form of <subject, relationship, object>. Finally, these triples are integrated into a structured knowledge graph to form a semantic network containing medical expertise. In addition, the knowledge graph can be stored in a graph database. The graph database can be Neo4j, JanusGraph or Amazon Neptune. Among them, the Neo4j graph database can enhance the expression of relationships, provide a complete graph reasoning language, support various graph mining algorithms, and use the Neo4j graph database to store the knowledge graph, which can effectively utilize the data expression, storage and reasoning centered on associated data in the graph database. These databases can efficiently manage node and edge relationships and support complex graph queries.

[0057] like Figure 3 The present invention provides a medical knowledge graph construction and reasoning system, which is used to implement the above-mentioned medical knowledge graph construction and reasoning method, and includes the following modules:

[0058] Knowledge graph construction module 101: obtain the content to be queried and the constructed medical knowledge graph.

[0059] Reasoning and query module 102: Based on the content to be queried, the constructed medical knowledge graph is identified and inferred through the graph neural network to determine the query result, which includes: the entity relationship of the content to be queried in the medical knowledge graph.

[0060] User interface interaction module 103: provides users with an intuitive graphical interface to reason about graphs based on user needs, explore medical concepts and their relationships; implements a visual display of user reasoning, such as graphical representation of entities and dynamic display of relationship paths, provides reasoning history and personalized recommendation functions; receives user reasoning requests, and displays reasoning results and graph exploration paths.

[0061] API development and integration module 104: Build API interfaces to allow external systems and applications to access and interact with graph data programmatically; implement API authentication, authorization, and request limiting to ensure data access security and system stability, and provide API documentation and developer tools; connect with system modules, process external API requests, interact with graph databases, and return inference or analysis results.

[0062] System integration and coordination module 105: ensures that all modules of the system work together, and is responsible for data flow and communication coordination between modules; among them, it implements synchronous and asynchronous communication mechanisms between modules, optimizes the data transmission process, reduces delays, and integrates logging and error handling mechanisms; it connects with all modules of the system, serves as the central hub for data flow, and ensures that data is transmitted accurately and efficiently between modules.

[0063] System maintenance and update module 106: responsible for the continuous monitoring, maintenance and update of the system, and optimizing the system according to user feedback and technological development; including integrating the automated testing framework, regularly checking the system performance and functions, implementing the hot update mechanism, and allowing the system to be updated without interrupting the service; connecting with various modules of the system, collecting system operation data and user feedback for system upgrades and function enhancements.

[0064] It should be understood that the embodiments of the present invention can be implemented in hardware, software or a combination thereof. Software can be stored in a volatile or non-volatile storage device (regardless of whether the data is deletable or rewritable) like a ROM, in a memory like a RAM, a memory chip, a device or an integrated circuit, or in a storage medium on which data can be optically or magnetically recorded and from which data can be read by a machine (e.g., a computer), such as a compact disc (CD), a digital versatile disc (DVD), a disk, or a tape. Moreover, the embodiments of the present invention can be implemented in a computer or portable terminal having a controller and a memory, and the memory is an example of a non-temporary computer-readable storage medium suitable for storing one or more programs including commands for implementing the embodiments of the present invention. Therefore, the present invention includes a program having a code for implementing the device or method defined by the claims, and a machine-readable non-temporary storage medium storing the program. The program can be transferred electrically by a medium such as a communication signal transmitted via a wired or wireless connection, and the program and its equivalent are included in the present invention.

[0065] The device may receive a program from a program providing device connected via a cable or wirelessly and store the program. The program providing device may include a program containing commands for implementing the embodiments of the present invention, a memory for storing information required by the embodiments of the present invention, a communication module for communicating with the mobile device via a cable or wirelessly, and a controller for transmitting the program to the device automatically or upon request of the device.

[0066] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A medical knowledge graph construction and reasoning method, characterized in that: include: Obtain the content to be queried and the constructed medical knowledge graph; According to the content to be queried, the constructed medical knowledge graph is identified and inferred through the graph neural network to determine the query result, which includes: the entity relationship of the content to be queried in the medical knowledge graph.

2. The medical knowledge graph construction and reasoning method according to claim 1 is characterized in that: The process of building a medical knowledge graph includes: Access medical research literature data; Mapping the medical research literature data into an embedding vector through an embedding model, and then performing bidirectional semantic feature learning on the embedding vector through an encoder to obtain a dynamic word vector; The decoder is used to parse the dynamic word vector that represents the ambiguity of words, capture the long-distance dependencies, temporal correlations and global features between data elements, extract entities based on the long-distance dependencies, temporal correlations and global features between data elements, and determine entity relationship triplets, thereby realizing the construction of a medical knowledge graph.

3. The medical knowledge graph construction and reasoning method according to claim 2 is characterized in that: The embedding model is a model trained by a partial word masking strategy, and the training process includes: The medical research literature data is segmented to obtain multiple word fragments; Randomly selecting a preset proportion of word fragments from multiple word fragments for masking to obtain masked word fragments and unmasked word fragments; Map the unmasked word fragments into embedding vectors and calculate the contextual information of the embedding vectors through the self-attention mechanism; Predicting the mask word fragment according to the context information to obtain a prediction result; Adjust the parameters of the embedded model according to the prediction results until the prediction accuracy of the embedded model meets the preset requirements.

4. The medical knowledge graph construction and reasoning method according to claim 1, characterized in that: The constructed medical knowledge graph is stored in a graph database, and the medical knowledge graph is queried through the graph database.

5. The medical knowledge graph construction and reasoning method according to claim 1, characterized in that: The recognition and reasoning of the constructed medical knowledge graph through the graph neural network includes: The entities and relationships in the knowledge graph are mapped into a graph data structure through a graph neural network to determine the entity relationships in the constructed medical knowledge graph, where the nodes in the graph data structure represent entities and the edges represent the relationships between entities.

6. The medical knowledge graph construction and reasoning method according to claim 5, characterized in that: The graph neural network includes a graph convolutional network, which performs feature learning on the graph data structure and infers the entity relationship in the medical knowledge graph, which is expressed as: Among them, H (l) H represents the node feature matrix of the lth layer, indicating the attributes of the entity corresponding to the node or its relationship with other entities; (l+1) Represents the node feature matrix of the l+1th layer, indicating the update of node features after a graph convolution operation; Represents the adjacency matrix A of the graph plus the identity matrix I N , used to represent the direct relationship between entities in the medical knowledge graph; for The diagonal matrix of the degree matrix is ​​used for normalization; W (l) represents the weight matrix of the lth layer, which is learned through the training process and used to update the node features; σ is the activation function, which is used to introduce nonlinearity to represent the node update process or the complex relationship between nodes.

7. The medical knowledge graph construction and reasoning method according to claim 1, characterized in that: Also includes: The query results and graph exploration paths are displayed on the user interface.

8. A medical knowledge graph construction and reasoning system, characterized in that: include: Knowledge graph construction module: obtain the content to be queried and the constructed medical knowledge graph; Reasoning and query module: Based on the content to be queried, the constructed medical knowledge graph is identified and inferred through the graph neural network to determine the query results. The query results include: the entity relationship of the content to be queried in the medical knowledge graph.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the medical knowledge graph construction and reasoning method described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of the medical knowledge graph construction and reasoning method described in any one of claims 1 to 7 are implemented.

Images