Research On Analytical Methods For Complex Medical Events

In the course of life,health is always the priority,but the disease is inevitable along the way,thence everyone has an indissoluble bond with medical care from birth.At the same time,the medical disputes and the ever-spreading antagonism between doctors and patients that have long plagued countries all over the world put forward higher requirements for evidence-based medical research.However,clinical decision-making implies a mystery.The efficient use of medical knowledge to extract medical events is extremely beneficial to the early detection and predictive intervention of clinical critical illnesses,and it can effectively alleviate the conflicts between doctors and patients.In the face of the high fatality rate of clinical critical illness,there is a lack of effective and explicable extrapolation and prediction methods.The main sticking point is that the medical incidents contained in clinical decision-making are difficult to transform into medical knowledge.According to research the decision-making analysis of the electronic medical record text of clinical critical diseases can effectively open up the context of medical knowledge and link the heterogeneous,high-dimensional,and massive distributed medical data.Therefore,this research intends to target the analysis of complex medical events as the goal,and to reconstruct an interpretable medical knowledge graph based on the implications of medical event decision-making,use knowledge transfer methods to conduct research on medical event attribution graph extraction,and explore adaptive self-constructed medical complex event retrieval methods.This research focuses on the strategic transformation of "health-centered" in our country,lay the foundation for the strategic needs of "healthy aging",and provide scientific and technological support for intelligent medical care.The main research contents and achievements of the thesis are as follows:(1)Reconstruction of the interpretable medical knowledge graph contained decision-makingBased on the requirement of medical event extraction and retrieval in decision-making,in view of the heterogeneous,high-dimensional,and massive distribution characteristics of medical data,we study the mechanism of multi-source aggregation,data sharing,and fusion linking.Based on the fusion of biomedical mechanism and expert domain experience knowledge with existing graphs,we realize the automatic generation of causal medical prototype knowledge graphs;For the transformation of clinical data contained in decision-making,clinical diagnosis and treatment guidelines and consensus,high-quality medical literature,etc.into medical knowledge,we have studied the adaptive perception method of medical knowledge based on multi-type data processing and deep support of domain knowledge.We have realized a multi-dimensional and multi-modal panoramic presentation of an interpretable medical knowledge graph guided by the implications of medical event decision-making.(2)Extraction method of medical event attribution graph based on knowledge transferBased on the spatio-temporal multidimensional extraction mechanism for unstructured medical data,we analyze the decision-making implications of the electronic medical records of patients admitted to the hospital for multiple times,and extract characteristic indicators in the time dimension;Feature indicators are extracted in the spatial dimension based on the knowledge transfer characteristics of the same background domain.According to the feature indicators of two dimensions,the method of joint extraction of entities and related knowledge in medical semantics is studied based on the principle of data fusion;With the support of the spatio-temporal data cube,we construct a deep medical event attribution graph based on joint analysis research.On the basis,we study the attribution graph support strategy to realize "autonomous perception-autonomous learning-autonomous transfer" in the whole link of medical events.(3)Self-adaptive and self-constructed medical complex event retrieval methodWe proposed the natural language description to structured query transformation(Text2SQL)method for the self-service retrieval of structured medical data;We study the complexity partition mechanism of multiple semantic knowledge retrieval based on the query syntax characteristics of the medical event;We adopt the pre-training mechanism to study the method of filling the semantic slot when data is missing with the support of medical knowledge atlas;Based on the learning theory of semantic network generalization,we design a natural language analysis and retrieval method with self-adaptive and self-constructed features to realize intelligent interaction of key medical events in the diagnosis and treatment process.(4)Prediction model for clinically critical medical eventsFocusing on critical medical events in an open,dynamic,and real environment,we verify and optimize the interpretable knowledge induction and migration techniques contained in decision-making;We optimize cross-modal fusion and human-machine division of labor for the modeling of uncertain clinical reasoning models;We do focus on model prediction and verification for the diagnosis and treatment of clinical critical illnesses,and improve the efficiency,accuracy,and precision of decision-making assistance in clinical medical events;the research results can be used to discover and reveal the rules and characteristics of disease evolution in the field of critical care medicine,support clinical experts' auxiliary diagnosis,and provide key technical support for evidence-based medicine research.