Research On Medical Entity Link

With the development of digital healthcare,electronic health data are at exponential increase.How to make full use of huge electronic health data to improve medical service is in hotspot.Clinical term mentions are usually of many variations in hospital information systems(HIS),which seriously impedes clinical data integration and usage.Clinical term normalization(i.e.,clinical term linking),which maps various clinical term mentions into standard clinical concepts,is a preliminary step of clinical data integration and usage,and a fundamental task of clinical information processing.There have been a large number of clinical term normalization methods proposed for electronic medical records(EMRs)in English,but few of methods proposed for EMRs in Chinese due to a lack of publically available datasets.Entity link method can be divided into two categories.one is single entity link,often appears in the outpatient records,without context information,what we usually deal with method such as string-match or rule-based.the other one is relational entity link,these entities are often among electronic medical records,with context information.In this paper,To the normalization of medical record in Chinese,for the disease among outpatient data,we proposed two disease normalization methods based on edit distance solving single entity link,with considering of the characteristics of different forms of them.Also,for the disease among EHRs,first we use conditional random field(CRF),to do clinical medical named entity recognition,then use ranking to finish entity link with medical context information.In order to evaluate these methods,we selected some records from a first-class hospital at grade 3 in China,use ICD-10(International Classification of Diseases and Related Health Problems,10 th revision)codes as link database,and recruited medical experts to manually annotate these records.Experiments on this dataset show that the proposed single entity link method is significantly better than traditional method at record-level and mention-level.For the relational entity link of EHRs,the ranking method we use also does well.