CN-121983225-A - Personalized medicine recommendation system based on dynamic graph neural network

CN121983225ACN 121983225 ACN121983225 ACN 121983225ACN-121983225-A

Abstract

The invention discloses a personalized medicine recommendation system based on a dynamic map neural network, which relates to the technical field of machine learning, and comprises a dynamic medical map construction module, a dynamic medical map generation module and a dynamic medical map generation module, wherein the dynamic medical map construction module is used for establishing a dynamic medical map; the dynamic learning result acquisition module is used for learning by using a dynamic graph neural network to generate a dynamic learning result, the dynamic embedded identification acquisition module is used for capturing time sequence change characteristics to generate a patient dynamic embedded identification, the recommended medicine list acquisition module is used for generating a recommended medicine list, and the medication recommendation suggestion acquisition module is used for generating medication recommendation suggestions. The invention solves the technical problems that the traditional medicine recommendation method cannot fully consider the dynamic change of the medical state of a patient along with time and is difficult to effectively correlate the dynamic relationship among related medical entities, achieves personalized medicine recommendation which is accurately attached to the real-time diagnosis and treatment requirements of the patient, continuously optimizes the recommendation result through a dynamic updating mechanism, and improves the technical effects of medicine recommendation accuracy and timeliness.

Inventors

YANG WANTING
ZHAO JUANJUAN
QIANG YAN
ZHAO JUN

Assignees

太原理工大学

Dates

Publication Date: 20260505
Application Date: 20251208

Claims (9)

1.A personalized medication recommendation system based on a dynamic graph neural network, the system comprising: the dynamic medical chart construction module is used for carrying out association analysis on the patient medical time sequence data set to construct a dynamic medical chart; The dynamic learning result acquisition module is used for learning the dynamic medical map by adopting the dynamic map neural network to generate a dynamic learning result; the dynamic embedded identification acquisition module is used for capturing time sequence change characteristics of the medical state of the patient based on the dynamic learning result and generating a dynamic embedded identification of the patient; The recommended medicine list acquisition module is used for carrying out personalized medicine recommendation on the patient according to the dynamic embedded identification of the patient and the medicine attribute information to generate a recommended medicine list; And the medication recommendation suggestion acquisition module is used for updating the dynamic medical map based on the recommended medication list, generating a dynamic medical optimization map, re-executing real-time medication recommendation analysis and generating medication recommendation suggestions.
2. The personalized medicine recommendation system based on a dynamic graph neural network of claim 1, wherein the process of constructing the patient medical time series data set comprises the following steps: The history diagnosis and treatment record acquisition unit is used for carrying out diagnosis and treatment analysis based on the electronic health record and extracting a history diagnosis and treatment record; The initial medical data set acquisition unit is used for carrying out characteristic standardization processing on the historical diagnosis and treatment record to generate an initial medical data set; The time sequence data segment acquisition unit is used for setting a time window, dividing the initial medical data set into data according to the time window, and generating a plurality of time sequence data segments; The medical feature vector acquisition unit is used for carrying out time sequence alignment processing on the plurality of time sequence data fragments, carrying out multi-time point analysis according to the time sequence of the historical diagnosis and treatment record according to the alignment result, and extracting medical feature vectors at a plurality of time points; and the patient medical time sequence data set acquisition unit is used for time sequence integration of medical feature vectors at a plurality of time points and constructing the patient medical time sequence data set.
3. The personalized medicine recommendation system based on the dynamic graph neural network of claim 1, wherein the association analysis is performed on the patient medical time series data set to construct the dynamic medical graph, and the system comprises: A plurality of node acquisition units that construct a plurality of nodes based on the patient medical time series data set, the plurality of nodes including a plurality of relationship edges; The initial static medical image acquisition unit is used for carrying out association connection on the plurality of nodes according to the plurality of relation edges to construct an initial static medical image; The time sequence relation edge acquisition unit is used for calling time information of the patient medical time sequence data set, mapping the time information to the initial static medical chart and adding time stamps to the relation edges to construct a plurality of time sequence relation edges; And the dynamic medical chart acquisition unit is used for expanding and updating the initial static medical chart according to the plurality of time sequence relation edges to construct the dynamic medical chart.
4. A personalized medicine recommendation system based on a dynamic graph neural network of claim 3, wherein a plurality of nodes are constructed based on the patient medical time series dataset, the plurality of nodes comprising a plurality of relational edges, the system comprising: A main body information acquisition subunit that analyzes the patient medical time series data set to determine a plurality of main body information, wherein the plurality of main body information includes patient information, disease information, drug information, and symptom information; An information node acquisition subunit that sets a patient node based on the patient information, sets a disease node based on the disease information, sets a drug node based on the drug information, and sets a symptom node based on the symptom information; a patient-disease relationship edge acquisition subunit that obtains a patient-disease relationship edge based on the patient node being associated with the disease information; a patient-drug relationship edge acquisition subunit that obtains a patient-drug relationship edge based on the association of the patient node with the drug node; a patient-symptom relationship edge acquisition subunit that acquires a patient-symptom relationship edge based on the association of the patient node with the symptom node; a disease-drug relationship edge acquisition subunit that obtains a disease-drug relationship edge based on the association of the disease node with the drug node; And a disease-symptom relationship edge acquisition subunit, which is used for acquiring a disease-symptom relationship edge based on the association of the disease node and the symptom node.
5. The personalized medicine recommendation system based on a dynamic graph neural network of claim 1, wherein the dynamic medical graph is learned by the dynamic graph neural network to generate a dynamic learning result, and the system comprises: The dynamic graph neural network construction unit is used for constructing a dynamic graph neural network comprising a space-time graph convolution module, wherein the dynamic graph neural network comprises a space convolution sub-layer and a time convolution sub-layer; the spatial feature acquisition unit is used for carrying out feature analysis on the dynamic medical map through the spatial convolution sub-layer and extracting spatial features; The time feature acquisition unit is used for carrying out feature analysis on the dynamic medical map through the time convolution sub-layer and extracting time features; The node space dependency relationship acquisition unit traverses a plurality of nodes of the dynamic medical graph based on the space characteristics to analyze, so as to acquire node space dependency relationships; the node time sequence evolution rule acquisition unit traverses a plurality of nodes of the dynamic medical graph based on the time characteristics to analyze, and obtains a node time sequence evolution rule; and the dynamic learning result acquisition unit is used for carrying out fusion learning according to the node space dependency relationship and the node time sequence evolution rule to acquire the dynamic learning result.
6. The personalized medicine recommendation system based on a dynamic map neural network of claim 5, wherein the plurality of nodes of the dynamic medical map are traversed based on the spatial features for analysis to obtain node spatial dependency relationships, the system comprising: The static image snapshot obtaining subunit is used for carrying out time slicing on the dynamic medical image based on the spatial characteristics to obtain a plurality of static image snapshots; The neighbor node aggregation information acquisition subunit is used for mapping the plurality of nodes to the plurality of static graph snapshots to execute graph convolution aggregation to acquire neighbor node aggregation information; the association strength obtaining subunit is used for carrying out association calculation on the plurality of nodes based on the neighbor node aggregation information to obtain a plurality of association strengths; And the node space dependency relationship acquisition subunit is used for setting propagation weights according to the plurality of association strengths, carrying out multi-layer convolution by combining the neighbor node aggregation information based on the propagation weights, and capturing the node space dependency relationship.
7. The personalized medicine recommendation system based on a dynamic graph neural network of claim 1, wherein the time sequence change characteristics of the medical state of the patient are captured based on the dynamic learning result, and the patient dynamic embedded identification is generated, and the system comprises: The patient node feature matrix acquisition unit is used for carrying out feature analysis on the patient nodes based on the dynamic learning result and extracting a patient node feature matrix; the time sequence data acquisition unit is used for analyzing based on the node characteristic matrix of the patient and extracting time sequence data; An attention score obtaining unit, configured to perform attention calculation on a patient node feature matrix according to the time series data, to obtain attention scores of a plurality of time node feature vectors; the attention weight distribution parameter acquisition unit is used for carrying out normalization processing according to the attention scores of the plurality of time node feature vectors to generate attention weight distribution parameters; a time sequence change feature acquisition unit for capturing time sequence change features of the medical state of the patient based on the dynamic learning result; The medical state change information acquisition unit is used for carrying out weighted summation on the time sequence change characteristics based on the attention weight distribution parameters to acquire medical state change information; And the dynamic embedded identification acquisition unit is used for carrying out multi-scale feature fusion according to the medical state change information to generate the dynamic embedded identification.
8. The personalized medicine recommendation system based on a dynamic graph neural network of claim 1, wherein personalized medicine recommendation is performed on a patient according to the patient dynamic embedding identification in combination with the medicine attribute information, and a recommended medicine list is generated, the system comprising: The drug attribute information acquisition unit is used for constructing a drug knowledge graph, analyzing drug attributes based on the drug knowledge graph and extracting drug attribute information; The multi-mode interaction matching result acquisition unit is used for carrying out matching calculation based on the dynamic patient embedding identification and the drug attribute information to generate a multi-mode interaction matching result; The candidate medicine information acquisition unit is used for carrying out medicine analysis on the patient according to the multi-mode interaction matching result and extracting a plurality of candidate medicine information; A treatment fitness score acquisition unit that performs a fit evaluation with the patient based on the plurality of candidate drug information, and obtains a treatment fitness score; The target candidate medicine information acquisition unit is used for judging based on the treatment fitness score and a fitness threshold, generating a screening instruction when the treatment fitness score is smaller than the fitness threshold, screening a plurality of candidate medicines through the screening instruction, and determining a plurality of target candidate medicine information; and a recommended drug list acquisition unit which is used for arranging the plurality of target candidate drug information in descending order according to the treatment suitability score and generating the recommended drug list.
9. The personalized medicine recommendation system based on a dynamic graph neural network of claim 4, wherein updating the dynamic medical graph based on the recommended medicine list generates a dynamic medical optimization graph to re-perform real-time medicine recommendation analysis, generates medicine recommendation suggestions, the system comprising: The association matching result acquisition unit is used for traversing the recommended medicine list and the medicine nodes to carry out association matching to generate an association matching result, wherein the association matching result is 1 or 0; The medicine matching node acquisition unit is used for considering that the recommended medicine list is successfully matched with the medicine nodes when the association matching result is 1, determining medicine matching nodes, and associating the recommended medicine list to the medicine matching nodes according to the association matching result; A new medicine node obtaining unit, configured to, when the association matching result is 0, consider that the matching between the recommended medicine list and the medicine node fails, and create a new medicine node; the patient-drug association relationship side construction unit is used for carrying out association analysis on the patient node, the drug matching node and the new drug node, and constructing a patient-drug association relationship side according to association coefficients; The recommended treatment relationship side construction unit is used for mapping the patient-drug association side to the dynamic medical chart for replacement to construct a recommended treatment relationship side; The patient embedded optimization identification acquisition unit updates the dynamic medical map based on the recommended treatment relationship, constructs the dynamic medical optimization map, re-executes dynamic map neural network analysis and generates a patient embedded optimization identification; And the medication recommendation suggestion acquisition unit is used for carrying out fine-grained optimization on the recommended medication list based on the patient embedded optimization identification and the real-time medical context information, so as to generate the medication recommendation suggestion.

Description

Personalized medicine recommendation system based on dynamic graph neural network Technical Field The invention relates to the technical field of machine learning, in particular to a personalized medicine recommendation system based on a dynamic graph neural network. Background The accurate medication recommendation has important significance for improving the treatment effect of patients and reducing adverse reactions of medication. In the prior art, personalized medication recommendation is based on static medical data matching, such as simple associated medication based on historical diagnosis and treatment records. The method plays a role in a data stabilization scene, but is exposed to limit when being applied to a dynamic diagnosis and treatment scene along with the upgrading of medical requirements. Because the medical state of the patient can change along with time, the traditional method cannot capture the time sequence change, and is difficult to effectively correlate the dynamic relationship among medical entities, the acquired medical information of the patient is incomplete, the recommended result is inaccurate, the real-time diagnosis and treatment requirements of the patient cannot be met, and the actual requirements of personalized medication are difficult to meet. Disclosure of Invention The application provides a personalized medication recommendation system based on a dynamic graph neural network, which solves the technical problems that the traditional medication recommendation method cannot fully consider the dynamic change of the medical state of a patient along with time and is difficult to effectively correlate the dynamic relationship among related medical entities. In view of the above problems, the application provides a personalized medicine recommendation system based on a dynamic graph neural network. The application provides a personalized medicine recommendation system based on a dynamic graph neural network, which comprises: The system comprises a dynamic medical image construction module, a dynamic medical image acquisition module, a medication recommendation suggestion acquisition module and a medication recommendation suggestion acquisition module, wherein the dynamic medical image construction module is used for carrying out association analysis on a patient medical time sequence data set to construct a dynamic medical image, the dynamic learning result acquisition module is used for learning the dynamic medical image by adopting the dynamic image neural network to generate a dynamic learning result, the dynamic embedding identification acquisition module is used for capturing time sequence change characteristics of a patient medical state based on the dynamic learning result to generate a patient dynamic embedding identification, the medication recommendation list acquisition module is used for carrying out personalized medication recommendation on a patient according to the patient dynamic embedding identification and combining with medication attribute information to generate a medication recommendation list, and the medication recommendation suggestion acquisition module is used for updating the dynamic medical image based on the medication recommendation list to generate a dynamic medical optimization image to re-execute real-time medication recommendation analysis to generate a medication recommendation suggestion. One or more technical schemes provided by the application have at least the following technical effects or advantages: According to the application, a patient medical time sequence data set is constructed and is subjected to association analysis to form a dynamic medical chart, a dynamic chart neural network is adopted to learn the dynamic medical chart to capture the time sequence change of the medical state of the patient and generate a dynamic embedded mark, the patient is matched and screened by combining with the medicine attribute information to generate a recommended medicine list, and the dynamic medical chart is updated based on a recommendation result to carry out re-optimization analysis, so that personalized medicine recommendation of the patient is realized, the medicine recommendation is more fit with the real-time diagnosis and treatment requirement of the patient, and the accuracy and timeliness are remarkably improved. The personalized medicine recommendation accurately fitting the real-time diagnosis and treatment requirements of the patient is achieved, the recommendation result is continuously optimized through a dynamic updating mechanism, and the technical effects of medicine recommendation accuracy and timeliness are improved. Drawings In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the pre