Medication mode mining system and method based on graph representation learning and RNN algorithm
The medication pattern mining system based on graph representation learning and RNN algorithm solves the problem of insufficient drug combination analysis in TCM medication pattern research, and realizes in-depth analysis of the relationship between TCM syndromes and drugs and intelligent medication pattern reasoning.
Patent Information
- Application Number
- CN202411377633.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-07
AI Technical Summary
Existing technologies are insufficient in the study of TCM medication patterns, especially in the analysis of the effects of drug combinations. They lack in-depth understanding of the details of medication and fail to effectively explore the complex relationship between drugs and patients' syndrome types and symptoms.
A medication pattern mining system based on graph representation learning and RNN algorithm is adopted. The system models the medication description module of patient characteristics and syndrome type into a multi-relationship heterogeneous undirected graph. Combined with graph fusion, attention and feedforward neural network modules, it automatically analyzes the direct and indirect relationship between TCM syndrome type and drug and forms the inference of the relationship between symptoms-syndrome type-drug.
It has enabled in-depth analysis and mining of TCM medication patterns, forming a TCM intervention and treatment support system based on big data models, and improving the intelligence and accuracy of medication patterns.
Smart Images
Figure CN119274822A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of medical service technology, and in particular relates to a medication pattern mining system and method based on graph representation learning and RNN algorithm. Background Art
[0002] Graph representation learning, also known as graph embedding, aims to represent nodes in the network as low-dimensional, real-valued, dense vector forms, so that the obtained vector forms can have the ability to represent and reason in the vector space, and can be easily and conveniently used as input to machine learning models. The obtained vector representation can then be applied to downstream computing tasks such as visualization, node classification, link prediction, and community discovery.
[0003] Recurrent Neural Network (RNN) is a type of recurrent neural network that takes sequence data as input, recursively in the direction of sequence evolution, and all nodes (recurrent units) are connected in a chain. Among them, Bidirectional Recurrent Neural Network (Bi-RNN) and Long Short-Term Memory Network (LSTM) are common recurrent neural networks.
[0004] Previous studies on TCM medication patterns have focused on the effects of drug combinations, but the details of medication use have often been insufficient. In TCM, there are more complex combination rules for the usage and dosage of a certain drug, the relationship between the four properties and five flavors and other drugs, and the matching with the patient's syndrome and symptoms. Summary of the invention
[0005] The purpose of the present invention is to solve the above-mentioned problems and to propose a medication pattern mining system and method based on graph representation learning and RNN algorithm.
[0006] On the one hand, in order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a medication pattern mining system based on graph representation learning and RNN algorithm, which comprises:
[0007] The data collection module connects the full-dimensional diagnostic information chest pain and heart pain disease database data;
[0008] A patient feature representation module, which obtains information from the data acquisition module and models the set of medical judgments involved in the patient diagnosis process as a multi-relation heterogeneous undirected graph;
[0009] The syndrome-type medication expression module takes the prescription as the central node, and the medicinal taste and drug characteristics as the surrounding nodes, and is modeled as a multi-relation heterogeneous undirected graph;
[0010] A graph fusion module, which automatically extracts features from the patient feature representation module and the syndrome type medication representation module, forms a distributed, low-dimensional, dense representation vector in space, and obtains semantic representation information of medication in low dimensions from both spatial and temporal dimensions;
[0011] A graph attention module, which learns a distributed representation of the semantic representation information of each node in the graph fusion module;
[0012] A feedforward neural network module, which obtains information from the graph attention module, uses the feedforward neural network to further extract patient characteristics in a new low-dimensional space, uses a recurrent neural network to use the correlation between TCM syndromes and various types of Chinese medicines involved in Chinese medicine prescriptions as important training data, and uses the recursive process of chain neuron connection to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and analyzes and derives the complex relationship between medication and medication effect through artificial intelligence to form an inference of the correlation relationship between symptoms-syndrome-drugs;
[0013] The medication pattern thrust module obtains the association relationship inference of the feedforward neural network module and realizes the medication pattern reasoning of the corresponding patient through a multi-label classification algorithm.
[0014] As a further description of the above technical solution:
[0015] In the feedforward neural network module, the recurrent neural network is a bidirectional recurrent neural network or a long short-term memory network.
[0016] On the other hand, in order to achieve the above-mentioned purpose, the present invention adopts the following method: a method for mining medication patterns based on graph representation learning and RNN algorithm, which includes the following method:
[0017] 1) The data collection module connects the full-dimensional diagnostic information chest pain and heart pain disease database data;
[0018] 2) The patient feature description module obtains information from the data collection module and models the medical judgment set involved in the patient diagnosis process as a multi-relation heterogeneous undirected graph. The syndrome type medication description module uses the prescription as the central node, and the taste and characteristics of the medicine as the surrounding nodes, and models it as a multi-relation heterogeneous undirected graph;
[0019] 3) The graph fusion module automatically extracts the features of the patient feature representation module and the syndrome type medication representation module, forming a distributed, low-dimensional, dense representation vector in space, and obtains the semantic representation information of medication in low dimensions from the dual dimensions of space and time;
[0020] 4) The graph attention module learns the distributed representation of the semantic representation information of each node in the graph fusion module;
[0021] 5) The feedforward neural network module obtains information from the graph attention module, uses the feedforward neural network to further extract patient characteristics in the new low-dimensional space, uses the recurrent neural network to use the correlation between TCM syndromes and various types of Chinese medicine involved in Chinese medicine prescriptions as important training data, and uses the recursive process of chain neuron connection to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and analyzes and derives the complex relationship between medication and medication effect through artificial intelligence, forming an inference of the correlation between symptoms, syndromes and drugs;
[0022] 6) The medication pattern thrust module obtains the association relationship inference of the feedforward neural network module and realizes the medication pattern reasoning of the corresponding patient through a multi-label classification algorithm.
[0023] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0024] 1. In the present invention, the information in the data acquisition module is obtained through the patient feature description module and the syndrome type medication description module to perform multi-relation heterogeneous undirected graph modeling, the graph fusion module automatically extracts the features of the patient feature description module and the syndrome type medication description module, the graph attention module and the feedforward neural network module adopt an end-to-end graph representation-based learning method, and at the same time make it possible to analyze and mine more complex medication patterns.
[0025] 2. In the present invention, a recurrent neural network is used to take the correlation between TCM syndromes and various types of TCM involved in TCM prescriptions as important training data, and a recursive process of chain neuron connection is used to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and the complex relationship between medication and medication effect is analyzed and derived by artificial intelligence. Finally, a core TCM medication inference engine that infers the correlation between symptoms, syndromes and drugs is formed as a big data model to support the medication category dimension in the process of TCM intervention treatment for UA patients. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 A framework for drug use pattern mining system based on graph representation learning and RNN algorithm Figure 1 .
[0027] Figure 2 A framework for drug use pattern mining system based on graph representation learning and RNN algorithm Figure 2 .
[0028] Figure 3 It is a multi-relation heterogeneous undirected graph of the patient feature description module in the medication pattern mining system based on graph representation learning and RNN algorithm.
[0029] Figure 4It is a multi-relation heterogeneous undirected graph of the syndrome type medication expression module in the graph fusion module of the medication pattern mining system based on graph representation learning and RNN algorithm.
[0030] Figure 5 This is a flowchart of a medication pattern mining method based on graph representation learning and RNN algorithm. DETAILED DESCRIPTION
[0031] 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.
[0032] Embodiment 1:
[0033] The S01 data collection module connects to the full-dimensional diagnostic information chest pain and heart pain disease database data;
[0034] The S02 patient feature description module obtains information from the data collection module and models the medical judgment set involved in the patient diagnosis process as a multi-relation heterogeneous undirected graph. The syndrome type medication description module uses the prescription as the central node, and the taste and characteristics of the medicine as the surrounding nodes, and models it as a multi-relation heterogeneous undirected graph.
[0035] The S03 graph fusion module automatically extracts the features of the patient feature description module and the syndrome type medication description module, forming a distributed, low-dimensional, dense representation vector in space, and obtains the semantic representation information of medication in low dimensions from the dual dimensions of space and time;
[0036] The S04 graph attention module learns the distributed representation of the semantic representation information of each node in the graph fusion module;
[0037] The S05 feedforward neural network module obtains information from the graph attention module, uses the feedforward neural network to further extract patient characteristics in the new low-dimensional space, and uses the recurrent neural network to use the correlation between TCM syndromes and various types of Chinese medicine involved in Chinese medicine prescriptions as important training data. The recurrent neural network is a bidirectional recurrent neural network, which uses the recursive process of chain neuron connection to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and analyzes and derives the complex relationship between medication and medication effect through artificial intelligence, forming an inference of the correlation between symptoms, syndromes and drugs;
[0038] The S06 medication pattern thrust module obtains the association relationship inference of the feedforward neural network module and realizes the medication pattern reasoning of the corresponding patient through a multi-label classification algorithm.
[0039] Embodiment 2:
[0040] The S01 data collection module connects to the full-dimensional diagnostic information chest pain and heart pain disease database data;
[0041] The S02 patient feature description module obtains information from the data collection module and models the medical judgment set involved in the patient diagnosis process as a multi-relation heterogeneous undirected graph. The syndrome type medication description module uses the prescription as the central node, and the taste and characteristics of the medicine as the surrounding nodes, and models it as a multi-relation heterogeneous undirected graph.
[0042] The S03 graph fusion module automatically extracts the features of the patient feature description module and the syndrome type medication description module, forming a distributed, low-dimensional, dense representation vector in space, and obtains the semantic representation information of medication in low dimensions from the dual dimensions of space and time;
[0043] The S04 graph attention module learns the distributed representation of the semantic representation information of each node in the graph fusion module;
[0044] The S05 feedforward neural network module obtains information from the graph attention module, uses the feedforward neural network to further extract patient characteristics in the new low-dimensional space, and uses the recurrent neural network to use the correlation between TCM syndromes and various types of Chinese medicine involved in Chinese medicine prescriptions as important training data. The recurrent neural network is a long-term and short-term memory network that uses the recursive process of chain neuron connections to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and analyzes and derives the complex relationship between medication and medication effect through artificial intelligence, forming an inference of the correlation between symptoms, syndromes, and drugs;
[0045] The S06 medication pattern thrust module obtains the association relationship inference of the feedforward neural network module and realizes the medication pattern reasoning of the corresponding patient through a multi-label classification algorithm.
[0046] Through data combing and algorithm practice, we summarize the following experiences and suggestions:
[0047] 1. During data combing and cleaning:
[0048] (1) Data standardization and normativeness:
[0049] It mainly involves the external information logic associated with the data model, such as: gender, occupation, disease, traditional Chinese medicine and other related rule definitions.
[0050] It would be best to form a set of digital standards and specifications for traditional Chinese medicine, which would be more conducive to digital and intelligent development.
[0051] (2) Data accuracy and stratification:
[0052] For digitization or data analysis, the accuracy and hierarchical classification of data are very important. These data should be summarized and refined based on the analysis of a large number of patients by TCM experts;
[0053] For example, which data is important and which data is secondary; which data can be classified and staged, and which data needs to be split and refined. These are mainly related to the decision-making role of patients and treatment.
[0054] (3) Logic for the importance of data:
[0055] The logic of data importance is mainly aimed at a large category of data. It needs to be classified and divided into important and secondary logical categories, and then the data is filled in the corresponding locations according to the rules.
[0056] For example, regarding the patient's symptoms and basic information about the patient, which ones are important and which ones are not; the diagnostic data and the order of Chinese medicine prescriptions are similar;
[0057] 2. During algorithm processing:
[0058] (1) Data integrity and regularity
[0059] For machine learning algorithms, its logic is to find patterns and memories from a certain amount of data, form a data model, and then output some reference suggestions based on the model.
[0060] This requires a certain logic and repetition rate for the integrity and regularity of the data itself, otherwise the algorithm will not be able to generate additional understanding capabilities.
[0061] (2) Data Magnitude:
[0062] As a machine learning algorithm, the larger the amount of data, the closer the learned algorithm is to reference and usable; therefore, it is very important to accumulate digital data.
[0063] (3) Algorithm optimization:
[0064] As doctors gain a deeper understanding of their patients, treatment techniques and methods may improve, leading to changes in data and treatment methods, which in turn will lead to changes in algorithms.
[0065] It is this continuous optimization and development that can promote the progress of an industry and continue to gain social recognition.
[0066] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A medication pattern mining system based on graph representation learning and RNN algorithm, characterized by: include: The data collection module connects the full-dimensional diagnostic information chest pain and heart pain disease database data; A patient feature representation module, which obtains information from the data acquisition module and models the set of medical judgments involved in the patient diagnosis process as a multi-relation heterogeneous undirected graph; The syndrome-type medication expression module takes the prescription as the central node, and the medicinal taste and drug characteristics as the surrounding nodes, and is modeled as a multi-relation heterogeneous undirected graph; A graph fusion module, which automatically extracts features from the patient feature representation module and the syndrome type medication representation module, forms a distributed, low-dimensional, dense representation vector in space, and obtains semantic representation information of medication in low dimensions from both spatial and temporal dimensions; A graph attention module, which learns a distributed representation of the semantic representation information of each node in the graph fusion module; A feedforward neural network module, which obtains information from the graph attention module, uses the feedforward neural network to further extract patient characteristics in a new low-dimensional space, uses a recurrent neural network to use the correlation between TCM syndromes and various types of Chinese medicines involved in Chinese medicine prescriptions as important training data, and uses the recursive process of chain neuron connection to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and analyzes and derives the complex relationship between medication and medication effect through artificial intelligence to form an inference of the correlation relationship between symptoms-syndrome-drugs; The medication pattern thrust module obtains the association relationship inference of the feedforward neural network module and realizes the medication pattern reasoning of the corresponding patient through a multi-label classification algorithm.
2. According to claim 1, a medication pattern mining system based on graph representation learning and RNN algorithm is characterized in that: In the feedforward neural network module, the recurrent neural network is a bidirectional recurrent neural network or a long short-term memory network.
3. A method for mining medication patterns based on graph representation learning and RNN algorithm as claimed in any one of claims 1 to 2, characterized in that: The following methods are included: 1) The data collection module connects the full-dimensional diagnostic information chest pain and heart pain disease database data; 2) The patient feature description module obtains information from the data collection module and models the medical judgment set involved in the patient diagnosis process as a multi-relation heterogeneous undirected graph. The syndrome type medication description module uses the prescription as the central node, and the taste and characteristics of the medicine as the surrounding nodes, and models it as a multi-relation heterogeneous undirected graph; 3) The graph fusion module automatically extracts the features of the patient feature representation module and the syndrome type medication representation module, forming a distributed, low-dimensional, dense representation vector in space, and obtains the semantic representation information of medication in low dimensions from the dual dimensions of space and time; 4) The graph attention module learns the distributed representation of the semantic representation information of each node in the graph fusion module; 5) The feedforward neural network module obtains information from the graph attention module, uses the feedforward neural network to further extract patient characteristics in the new low-dimensional space, uses the recurrent neural network to use the correlation between TCM syndromes and various types of Chinese medicine involved in Chinese medicine prescriptions as important training data, and uses the recursive process of chain neuron connection to automatically analyze the direct relationship between syndromes and drugs and the indirect relationship through chain conduction, and analyzes and derives the complex relationship between medication and medication effect through artificial intelligence, forming an inference of the correlation between symptoms, syndromes and drugs; 6) The medication pattern thrust module obtains the association relationship inference of the feedforward neural network module and realizes the medication pattern reasoning of the corresponding patient through a multi-label classification algorithm.