Intelligent medical treatment system based on graph convolutional neural network model

Abstract

The invention discloses an intelligent medical treatment system based on a graph convolutional neural network model, and particularly relates to the field of medical care, and the system comprises a data collection and processing module, a first knowledge relation model construction module, a graph embedding module, a graph convolutional neural network module, a feature extraction and learning module, and an intelligent diagnosis and treatment recommendation module. The data acquisition and processing module is responsible for acquiring medical data and cleaning, fusing and preprocessing the medical data, and the first knowledge relationship model construction module establishes a first medical knowledge relationship model covering patients, diseases, symptoms, drug entities and relationships thereof; mapping entities and relationships in the first knowledge relationship model into low-dimensional vector representation according to a graph embedding module, extracting feature information in the medical data by using a graph convolutional neural network, and learning a patient condition, diagnosis and treatment association rule; and finally, an intelligent diagnosis and treatment recommendation module provides intelligent diagnosis and treatment scheme recommendation and medication suggestion services for the patient according to the learned features and relationships.

Description

Technical Field

[0001] The present invention relates to the field of medical care technology, and more specifically, to an intelligent medical treatment system based on a graph convolutional neural network model. Background Art

[0002] With the continuous development of medical technology and data science, smart medical systems are becoming more and more common. These smart medical systems use technologies such as big data, artificial intelligence, and machine learning to improve the efficiency and quality of medical services.

[0003] In recent years, graph convolutional neural networks have been shown to be an effective model for machine learning tasks on graph data. In the medical field, the relationship between patients and doctors can be viewed as a graph structure, in which nodes represent patients or doctors and edges represent the relationship between them.

[0004] Smart medical systems based on graph convolutional neural networks can use the information of this relationship graph to provide personalized diagnosis and treatment recommendations for doctors and patients. By analyzing the patient's medical history, symptoms, and doctor's expertise, and combining data from other patients, the system can generate the best treatment plan for each patient.

[0005] In the traditional medical treatment process, patients usually need to go to a medical institution, such as a hospital or clinic, in person, wait for the doctor's consultation and physical examination. The doctor will make a diagnosis and develop a treatment plan based on the patient's medical history, symptoms and signs.

[0006] However, there are still some shortcomings in its actual use. For example, in the traditional medical treatment process, patients cannot determine which medical institution to go to, and the communication between doctors and patients is based on face-to-face communication. Doctors need to rely on their own experience and expertise to provide patients with diagnosis and treatment advice. Therefore, an intelligent medical system based on the graph convolutional neural network model came into being. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a smart medical treatment system based on a graph convolutional neural network model to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] Data acquisition and processing module: used to collect medical data and perform data cleaning, fusion and preprocessing on the collected data;

[0010] First knowledge relationship model construction module: used to construct the first medical knowledge relationship model, including patients, diseases, symptoms, drugs and their relationships;

[0011] Graph embedding module: used to map entities and relationships in the first knowledge relationship model into low-dimensional vector representations for representing graph structure data;

[0012] Graph Convolutional Neural Network Module: used to represent, design and implement graph convolutional neural network models based on graph embedding;

[0013] Feature extraction and learning module: used to extract feature information from medical data using graph convolutional neural network models, and learn the relationship between patient conditions and identification and treatment;

[0014] Intelligent diagnosis and treatment recommendation module: used to provide patients with intelligent diagnosis and treatment plan recommendations and medication advice services based on learned features and relationships.

[0015] Preferably, in the data acquisition and processing module, the types of medical data collected include patient information, medical records, test results, and imaging data; the hospital cooperates with relevant medical institutions or medical information systems to obtain permission and authorization for medical data; and uses API interfaces and other methods to obtain medical data from different data sources, and ensure the timeliness and accuracy of data acquisition.

[0016] Preferably, in the first knowledge relationship model construction module, based on the knowledge system in the medical field, the entity types in the first knowledge relationship model are first determined, and the entity types include patients, diseases, symptoms, and drugs, and the relationships between different entities are defined, such as patients suffering from a certain disease, a certain drug treating a certain disease, etc.; based on the data source, natural language processing, relationship extraction and other technologies are used to extract the relationships between entities from text data; entities and relationships are represented as a graph structure, and the data is stored in a Neo4j graph database that supports graph query and graph analysis; information describing the same entity in different data sources is linked and uniformly represented to ensure that entities in different data sources can correspond to the same entity in the first knowledge relationship model.

[0017] Preferably, in the graph embedding module, the first knowledge relationship model is represented as a graph structure, and the relationship between entities is represented by constructing an adjacency matrix between entities; a graph representation method of an adjacency linked list is adopted.

[0018] Preferably, the graph convolutional neural network module prepares the medical first knowledge relationship model data, including entity, relationship and attribute information, and represents the graph data in the form of an adjacency matrix and a feature matrix;

[0019] A graph convolutional neural network model is designed to learn node representation on graph data; the graph convolutional neural network model updates the representation of the node by aggregating the features of the node and its neighboring nodes, thereby capturing the structural information between the nodes; and an importance score is assigned to each node.

[0020] Preferably, a graph convolutional neural network model is used to extract characteristic information of entities such as patients, diseases, symptoms, and drugs from the medical first knowledge relationship model, and learn the complex association patterns between them; by iteratively optimizing model parameters, feature representation can be learned on the training set, thereby extracting the patterns of association between the patient's condition and diagnosis and treatment.

[0021] Preferably, an intelligent diagnosis and treatment recommendation model is designed based on the learned feature representations and relationships, and methods such as rule-based recommendation systems, content-based recommendation systems, or collaborative filtering can be used.

[0022] Technical effects and advantages of the present invention:

[0023] The present invention uses a graph convolutional neural network to analyze and process the collected data to extract valuable information and patterns; based on the patient's condition and needs, the information on each hospital node is learned, and when the patient needs to see a department, the most suitable hospital is recommended to the patient based on the information classification on the node; based on the weights between the nodes and the patient's current location, the nearest medical location is recommended to the patient to reduce the patient's travel costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of system module connection of the present invention.

[0025] Figure 2 It is a schematic diagram of the method flow of the present invention.

[0026] Figure 3 It is the graph convolutional neural network model of the present invention.

[0027] Figure 4 A schematic diagram of the code for building the graph convolutional neural network of the present invention.

[0028] Figure 5 It is a schematic diagram of the iteration of the model of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0030] See also Figure 1As shown, the present invention relates to a smart medical treatment system based on a graph convolutional neural network model, which includes a data acquisition and processing module, a first knowledge relationship model construction module, a graph embedding module, a graph convolutional neural network module, a feature extraction and learning module, and an intelligent diagnosis and treatment recommendation module.

[0031] The data acquisition and processing module is connected to the first knowledge relationship model construction module, the first knowledge relationship model construction module is connected to the graph embedding module, the graph embedding module is connected to the graph convolutional neural network module, the graph convolutional neural network module is connected to the feature extraction and learning module, and the feature extraction and learning module is connected to the intelligent diagnosis and treatment recommendation module.

[0032] The data acquisition and processing module is used to collect medical data and perform data cleaning, fusion and preprocessing on the collected data;

[0033] In the data collection and processing module, the types of medical data collected include patient information, medical records, examination results, and imaging data; the hospital cooperates with relevant medical institutions or medical information systems to obtain permission and authorization for medical data; and uses API interfaces and other methods to obtain medical data from different data sources and ensure the timeliness and accuracy of data collection;

[0034] The data acquisition and processing module includes the following units:

[0035] Data cleaning unit: cleans the collected raw data, including removing duplicate data, processing missing values, arranging data formats, detecting and processing outliers, and identifying and processing abnormal data;

[0036] Data fusion unit: Integrate and fuse medical data from different data sources to build a unified data storage and management system; use data fusion technology to connect and associate data of different dimensions;

[0037] Data preprocessing unit: Perform data preprocessing, including data transformation, normalization, and standardization, to make the data meet the requirements of model training. Extract and select features from the data, select the most representative features, and prepare for subsequent model training and reasoning.

[0038] Data storage and management unit: Design appropriate data storage structure and management strategy to ensure the security and privacy of medical data; ensure the efficiency and scalability of the data storage system to support the storage and management needs of large-scale data.

[0039] The first knowledge relationship model construction module is used to construct a medical first knowledge relationship model, including patient, disease, symptom, drug entities and their relationships;

[0040] In the first knowledge relationship model construction module, based on the knowledge system in the medical field, the entity types in the first knowledge relationship model are first determined, and the entity types include patients, diseases, symptoms, and drugs, and the relationships between different entities are defined, such as patients suffering from a certain disease, a certain drug treating a certain disease, etc.; based on the data source, natural language processing, relationship extraction and other technologies are used to extract the relationship between entities from text data; entities and relationships are represented as a graph structure, and the data is stored in a Neo4j graph database that supports graph query and graph analysis; information describing the same entity in different data sources is linked and uniformly represented to ensure that entities in different data sources can correspond to the same entity in the first knowledge relationship model; the quality of the constructed medical first knowledge relationship model is evaluated, including evaluation of entity integrity, relationship accuracy, and relationship logic validity, to ensure the quality and usability of the first knowledge relationship model; the constructed medical first knowledge relationship model is presented using visualization tools to facilitate user understanding and query; at the same time, the first knowledge relationship model is applied to medical data analysis, knowledge reasoning, patient risk assessment and other fields to provide basic support for subsequent graph analysis.

[0041] The graph embedding module is used to map entities and relationships in the first knowledge relationship model into low-dimensional vector representations for representing graph structure data;

[0042] In the graph embedding module, the first knowledge relationship model is represented as a graph structure, and the relationship between entities is represented by constructing an adjacency matrix between entities; a graph representation method of an adjacency list is adopted;

[0043] The graph is represented as a set of vertices and edges, expressed as: G = (V, E), where V represents the vertex set and E represents the edge set. Suppose the number of vertices in the graph G is N, the number of edges is M, and a line connecting vertices (V i , V j ); where V = {v1, v2, v3, v4, v5}, E = {(v1, v2), (v1, v3), (v2, v4), (v2, v3), (v3, v4), (v4, v5)}; In order to recommend suitable hospitals and hospitals closest to patients, hospitals in China are taken as points, and each point is connected by an edge to form a directed graph;

[0044] The Node2Vec embedding learning algorithm is selected to map entities and relationships into low-dimensional vector representations, and then subsequent graph convolution operations are performed; based on the adjacency matrix of the first knowledge relationship model, the model is trained using the selected embedding learning algorithm to learn the low-dimensional vector representations of entities and relationships; the model parameters are optimized by minimizing the loss function so that entities and relationships capture the semantic relationship between them in the vector space.

[0045] The trained embedding model is evaluated, and the performance of the model is evaluated through tasks such as node classification and link prediction. At the same time, the trained embedding model is applied to subsequent graph convolutional network analysis tasks.

[0046] The graph convolutional neural network module is used to design and implement a graph convolutional neural network model based on graph embedding representation;

[0047] The graph convolutional neural network module prepares the medical first knowledge relationship model data, including entity, relationship and attribute information, and represents the graph data in the form of an adjacency matrix and a feature matrix; designs a graph convolutional neural network (graph convolutional neural network model) model for learning node representation on the graph data; the graph convolutional neural network model updates the representation of the node by aggregating the features of the node and its neighboring nodes, thereby capturing the structural information between the nodes;

[0048] Each node is assigned an importance score, which is calculated as follows:

[0049] Among them, F is the representation of all nodes, ∈ represents the activation function, which means that the adjacency matrix with self-connection is added, H represents the adjacency matrix, I represents the degree matrix, J represents the feature of the node, and Θatt∈RN×1 is the weight parameter;

[0050] Among them, F is the representation of all nodes, ∈ represents the activation function, and F T It is represented as a transposed matrix. Pooling operation can be performed according to the node importance score and topological structure, unimportant nodes are discarded, the adjacency matrix and node features are updated, and the pooling result is obtained.

[0051] The adjacency relationship between nodes is reconstructed and learned, and the calculation method of the graph autoencoder is defined as follows:

[0052] F = GNN (J, H), where F is the representation of all nodes, J is the feature of the node, and H is the adjacency matrix. The graph convolutional neural network model is used to encode and learn the attribute information and structural information of the graph at the same time;

[0053] The inner product of the vector is used to represent the adjacency relationship between nodes. The calculation method of the reconstruction loss function of the graph autoencoder is as follows:

[0054] Among them, L recon It is represented as the reconstruction loss function, and H is represented as the adjacency matrix;

[0055] Due to the problem of over-smoothing, graph neural networks can easily learn similar expressions for adjacent nodes, which causes the decoded adjacency matrix to quickly approach the original adjacency matrix H, and it is difficult to effectively optimize the model parameters. Therefore, in order to enable graph neural networks to learn more effective data distributed representations, like autoencoders, constraints are added to the loss function. The constraints include:

[0056] (1) Appropriately add random noise or perform zero processing on the feature matrix J of the original image data;

[0057] (2) Delete an appropriate proportion of edges from the adjacency matrix H of the original image data, or modify the weight values ​​on the edges;

[0058] The basic framework includes the inference model, the generation model, and the loss function; the inference model is the encoder and the generation model is the decoder;

[0059] The calculation method of the inference model is as follows:

[0060] Among them, F is the representation of all nodes, J is the feature of the node, and H is the adjacency matrix;

[0061] Among them, F is the representation of all nodes, J is the feature of the node, H is the adjacency matrix, diag is the diag function, and ε is the activation function;

[0062] The calculation method of the generative model is as follows:

[0063] F is the representation of all nodes, H is the adjacency matrix,

[0064] Among them, F is the representation of all nodes, H is the adjacency matrix, and ε is the activation function;

[0065] Through online data research, we found that each hospital has the most authoritative department, which is encoded as the label of the graph convolutional neural network;

[0066] When training the model network model, when the model iterates about 5000 times, the loss function value tends to be around 0.34.

[0067] Among them, the graph convolutional neural network model includes a training set, a validation set and a test set. The training set is used to train the graph convolutional neural network model, and the model parameters are optimized by minimizing the loss function; hyperparameters are tuned on the validation set, and the model performance is evaluated on the test set; the trained graph convolutional neural network model is applied to medical data to learn the complex relationships and characteristics between entities such as patients, diseases, and drugs, and is used for tasks such as prediction, classification, and clustering of medical data to assist medical decision-making and diagnosis; the graph convolutional neural network model is then evaluated, including evaluation of model performance indicators, interpretability analysis, etc., and the model is tuned and improved based on the feedback results.

[0068] The feature extraction and learning module is used to extract feature information from medical data using a graph convolutional neural network model, and learn the relationship between the patient's condition and identification and treatment;

[0069] In the feature extraction and learning module, the graph convolutional neural network model is used to extract the feature information of entities such as patients, diseases, symptoms, and drugs from the medical first knowledge relationship model, and learn the complex association rules between them; by iteratively optimizing the model parameters, the feature representation can be learned on the training set, thereby extracting the rules of association between the patient's condition and diagnosis and treatment;

[0070] Use the training set to train the graph convolutional neural network model, and optimize the model hyperparameters on the validation set; continuously optimize the model parameters through back-propagation algorithm and gradient descent to improve the performance and generalization ability of the model. Evaluate the performance of the trained and optimized graph convolutional neural network model on the test set; apply the model to medical decision-making tasks such as patient diagnosis, disease prediction, and drug recommendation, thereby providing data-driven support and decision-making basis for the medical field.

[0071] The intelligent diagnosis and treatment recommendation module is used to provide patients with intelligent diagnosis and treatment plan recommendations and medication advice services based on the learned features and relationships;

[0072] In the intelligent diagnosis and treatment recommendation module, an intelligent diagnosis and treatment recommendation model is designed based on the learned feature representations and relationships, and methods such as a rule-based recommendation system, a content-based recommendation system, or collaborative filtering can be used.

[0073] Provide patients with personalized treatment recommendations and medication advice based on patient medical history information, diagnosis results, medication history and other factors; use training data to train the recommendation model, and evaluate the performance of the model through cross-validation and other methods, including accuracy, recall rate, F1 value and other indicators; tune the model according to the evaluation results to improve the accuracy and practicality of the recommendation system; deploy an intelligent diagnosis and treatment recommendation system, integrate the model into the medical information system, and provide doctors and patients with treatment recommendations and medication advice services.

[0074] See also Figure 2 As shown, in this embodiment, it should be specifically explained that the present invention provides a smart medical treatment system based on a graph convolutional neural network model, comprising the following steps:

[0075] Step Q1: Collect medical data, and clean, fuse and preprocess the collected data;

[0076] Step Q2: Construct the first medical knowledge relationship model, including patient, disease, symptom, drug entities and their relationships;

[0077] Step Q3: Mapping entities and relationships in the first knowledge relationship model into low-dimensional vector representations for representing graph structure data;

[0078] Step Q4: Design and implement a graph convolutional neural network model based on graph embedding representation;

[0079] Step Q5: Use the graph convolutional neural network model to extract feature information from medical data and learn the relationship between the patient's condition and diagnosis and treatment;

[0080] Step Q6: Based on the learned features and relationships, provide patients with intelligent diagnosis and treatment plan recommendations and medication advice services.

[0081] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A smart medical treatment system based on a graph convolutional neural network model, characterized in that: include: Data acquisition and processing module: used to collect medical data and perform data cleaning, fusion and preprocessing on the collected data; The first knowledge relationship model construction module is used to construct the first knowledge relationship model, including patient, disease, symptom, drug entities and their relationships; Graph embedding module: used to map entities and relationships in the first knowledge relationship model into low-dimensional vector representations for representing graph structure data; Graph Convolutional Neural Network Module: used to represent, design and implement graph convolutional neural network models based on graph embedding; Feature extraction and learning module: used to extract feature information from medical data using graph convolutional neural network models, and learn the relationship between patient conditions and identification and treatment; Intelligent diagnosis and treatment recommendation module: used to provide patients with intelligent treatment plan recommendations and medication advice services based on learned features and relationships; In the data collection and processing module, the types of medical data collected include patient information, medical records, examination results, and imaging data; the hospital cooperates with relevant medical institutions or medical information systems to obtain permission and authorization for medical data; and uses API interfaces and other methods to obtain medical data from different data sources and ensure the timeliness and accuracy of data collection; The first knowledge relationship model building module determines the entity types in the first knowledge relationship model building module according to the knowledge in the medical field. The entity types include patients, diseases, symptoms, and drugs, and defines the relationships between different entities. In the first knowledge relationship model, the relationships between these entities can be defined as: Patient-disease relationship: represents the relationship between a patient and his or her disease, including the type of disease, medical history, etc. Disease-symptom relationships: Describe the various symptoms that a specific disease may cause. This information helps doctors diagnose the disease and recommend treatment options. Disease-drug relationship: Indicates which drugs can be used to treat specific diseases, including information such as the recommended dosage and frequency of use of the drugs; Patient-drug relationship: Record the patient's current or historical medication information to help doctors understand the patient's treatment history and drug allergies; These entity types and relationship definitions will serve as the basis for the first knowledge relationship model. The model constructed in this way can effectively organize and utilize complex information in the medical field, providing accurate data support and intelligent service functions for the smart medical system.

2. According to claim 1, a smart medical treatment system based on a graph convolutional neural network model is characterized by: In the graph embedding module, the first knowledge relationship model is represented as a graph structure, and the relationship between entities is represented by constructing an adjacency matrix between entities; a graph representation method of an adjacency list is adopted; A graph is represented as a set of vertices and edges, expressed as: A = (B, C), where B represents the vertex set and C represents the edge set. Suppose the number of vertices in graph A is N (D), the number of edges is M (E), and a line connecting vertices (B i , B j ); Among them, B = {B1, B2, B3, B4, B5}, E = {(B1B2), (B1, B3), (B2, B4), (B2, B3), (B3, B4), (B4, B5)}; In order to recommend suitable hospitals and hospitals closest to patients, hospitals in China are regarded as points, and each point is connected by edges to form a directed graph.

3. According to claim 1, a smart medical treatment system based on a graph convolutional neural network model is characterized by: The graph convolutional neural network module prepares the medical first knowledge relationship model data, including entity, relationship and attribute information, and represents the graph data in the form of an adjacency matrix and a feature matrix; Each node is assigned an importance score, which is calculated as follows: Among them, F is the representation of all nodes, ∈ represents the activation function, represents the addition of the self-connected adjacency matrix, H represents the adjacency matrix, I represents the degree matrix, J represents the node feature, Θatt∈RN×1 is the weight parameter; Among them, F is the representation of all nodes, ∈ represents the activation function, and F T Represented as a transposed matrix; pooling operation can be performed according to the node importance score and topological structure, unimportant nodes are discarded, the adjacency matrix and node features are updated, and the pooling result is obtained F = GNN (J, H), where F is the representation of all nodes, J is the feature of the node, and H is the adjacency matrix. The graph convolutional neural network model is used to encode and learn the attribute information and structural information of the graph at the same time; The inner product of the vector is used to represent the adjacency relationship between nodes; the calculation method of the reconstruction loss function of the graph autoencoder is as follows: Among them, L recon It is represented as the reconstruction loss function, and H is represented as the adjacency matrix; Due to the over-smoothing problem, graph neural networks can easily learn similar expressions for adjacent nodes, which causes the decoded adjacency matrix to quickly approach the original adjacency matrix H, and the model parameters are difficult to effectively optimize. Therefore, in order to enable graph neural networks to learn more effective data distributed representations, a constrained target is added to the loss function, just like the autoencoder.

4. According to claim 1, a smart medical treatment system based on a graph convolutional neural network model is characterized by: The basic framework includes an inference model, a generation model, and a loss function; wherein the inference model is an encoder and the generation model is a decoder; Through online data research, it is concluded that each hospital has the most authoritative department, which is encoded as the label of the graph convolutional neural network.

5. According to claim 1, a smart medical treatment system based on a graph convolutional neural network model is characterized by: In the feature extraction and learning module, a graph convolutional neural network model is used to extract feature information of entities such as patients, diseases, symptoms, and drugs from the first knowledge relationship model, the graph convolutional neural network model is trained using a training set, and the model hyperparameters are optimized on a validation set; the model parameters are continuously optimized through back-propagation algorithms and gradient descent and other methods to improve the performance and generalization ability of the model; the performance of the trained and optimized graph convolutional neural network model is evaluated on a test set; and the model is applied to medical decision-making tasks such as patient diagnosis, disease prediction, and drug recommendation, thereby providing data-driven support and decision-making basis for the medical field.

6. According to claim 1, a smart medical treatment system based on a graph convolutional neural network model is characterized by: In the intelligent diagnosis and treatment recommendation module, an intelligent diagnosis and treatment recommendation model is designed based on the learned feature representations and relationships.

Images