Development Of Computational Methods For Extracting Information From Chinese Electronic Medical Records

The rapid development of information technology promotes the hospital information construction.Besides,the national policy lays a solid basis for the construction of medical information system such as electronic medical record.Therefore,abundant medical data are brought out,and electronic medical record attracted a wide spread attention.Electronic medical record is the important clinical information resource which is generated during the curative activity.It also includes much medical knowledge which is closely related to the health state of patients.Extracting useful information from electronic medical record will greatly promote the development of medical service.In this thesis,we first introduced the research status of information extraction of electronic medical record.Then the overview of information extraction of electronic medical record,including the features of language and structure of electronic medical record,the main method of information extraction were described.The algorithm combining CRF and linguistic rules are designed for the named entity recognition from Chinese electronic medical record.And the method based on SVM algorithm was proposed for the study of the assertion classification of Chinese electronic medical record.The main works are shown as follows:(1)This thesis designed a method to achieve named entity recognition from Chinese electronic medical record which combines CRF and linguistic rules.First,aiming at the short of data in electronic medical record for the study,this paper collects 200 electronic medical record typical cases of internal medicine from internet,and labels the entity and the assertion types of entities based on the annotation guidelines "Electronic Medical Records in Chinese Named Entity and Entity Relationship”specified by Harbin Institute of Technology intelligence research laboratory.Then we proposed a method based on CRF algorithm and linguistic rules to achieve named entity recognition from Chinese electronic medical record.The experiment results suggest that our named entity recognition method is efficient with the best F-measure of 92.68%in the test set.(2)A method based on SVM algorithm was proposed for the assertion classification of Chinese electronic medical record.After analyzing the text and syntax features of Chinese medical records,we build a SVM-based classifier for Chinese named entity assertion classification based on the result of named entity recognition.The features of SVM classifier include context features,section headings features,parts of speech(POS)features and cues information features.Experimental results and analysis show that the proposed SVM-based assertion classifier achieved good performance with the best F-measure of 94.10%.(3)A system extracting information from Chinese electronic medical record is designed and implemented,which provides an experimental test platform and reference for the further research.Based on the named entity recognition and assertion classification models mentioned above,the system completes named entity recognition and assertion classification from Chinese medical recordsOverall,this thesis enables the information extraction and data mining of Chinese Medical Electronic Records,thus helping the effective usage of tremendous clinical data in China.