Research On Deep Medical Retrieval Model Based On Expansion Of Medical Knowledge Bases

Many studies have shown that consulting a large number of relevant medical literature can effectively help doctors to make more accurate judgments,reduce the occurrence of errors,and provide patients with the best treatments.However,due to a large number of biomedical literature on the network,it is an urgent problem that how to quickly and efficiently access the most relevant literature.Besides,the speed of renewal of clinicians’knowledge and methods cannot keep up with the rapid development of the latest medical ideas and technology,and the knowledge lag may cause considerable losses in the clinic.Scientific use of medical retrieval can help clinicians to keep abreast of the latest medical knowledge in their field.The purpose of the medical literature retrieval is to find useful information of current medical records from a large number of the published biomedical literature quickly and effectively,to reduce possible clinical errors and improve clinical quality.It can assist doctors in making clinical decision support and providing better treatment options for patients.The main task of this paper is to regard a given Electronic Medical Record(EMR)as a query,so as to retrieve the most relevant documents in a large number of biomedical documents,which can provide literature support for current medical records,and help clinicians make clinical decisions.Due to the particularity of medical retrieval,the effect of previous retrieval methods may not be good enough.The main reasons are as follows:Firstly,the information on electronic medical records may be too rough to reflect the actual information needs of patients.Secondly,the same symptoms in electronic medical records may point to entirely different diseases.Thirdly,there may be several different terms for the same disease concept.According to the past clinical experience,it will be helpful to improve the effect of information retrieval if we can pre-infer the patients’possible diseases,examinations and treatments based on medical knowledge before carrying out medical information retrieval.Regarding the issue above,this paper proposes a universal Deep Medical Retrieval Model(DMRM),which can expand queries based on relevant medical knowledge.In view of the problem of medical concepts mismatch which is caused by different expressions of the same medical concepts in different data sources,this paper introduces the Unified Medical Language System(UMLS)to process medical data,so as to determine the unique identification of medical concepts,and unify different expressions.Given an electronic medical record as query,the model uses MetaMap to extract medical concepts from the query.Then,the extension items are inferred from the medical-related public knowledge bases for query expansion.The expanded query is used to retrieve the Pseudo Relevance Feedback(PRF)set by BM25 model.The pseudo-relevance feedback document sets and the extended queries are applied to the neural network model to sort candidate documents,so as to obtain the final document ranking lists.In order to verify the retrieval effect,this paper uses the standard data sets of Text REtrieval Conference(TREC)Clinical Decision Support(CDS)project in 2014 and 2015.The experimental results prove that DMRM,a deep medical retrieval model based on the expansion of medical knowledge bases,has superior performance.Its retrieval effect is much higher than other existing information retrieval models.It is easier to implement than other models,and has a broader scope of application.