Design And Implementation Of Discharge Summaries Error Detection System Based On NER

With the development of the medical technology, higher requirements upon the quality as well as the safety of the medical facilities has been put forward, especially for the medical documents which widely catches the eye of the press and public. A lot of research has indicated that as a key element of the medical documents, there exists a series of problems including the inaccuracy, invalidity and omission of the information in discharge summaries. Since the errors in discharge summaries are mainly in form of narrative context, it cannot be avoided by optimizing the enclosure of the discharge summaries. In addition, today’s auto-correction technology is mainly focusing on detecting on the meaning of the words and phrases of the text, and thus it has difficulty in judging the completeness of discharge summaries. Under such circumstances, in this paper an discharge summaries error detection system based on named entity recognition is proposed.The basic architecture of such an error-detecting system consists of four parts: named entity recognition, error detection, correction suggestion and an integrated platform. Named entity recognition could help to extract important diagnosis information of the discharge summaries, where the F-measure could be achieved to 88.89% based on optimizing the results from four sub experiments. Error detection is achieved by proofreading the medical data with the predefined dictionary and for each error entity a corresponding correction suggestion is given based on the algorithm of fuzzy matching with the knowledge base. The integrated platform and database assure the data exchange between the electronic medical record system and the proposed system. The Discharge summaries proposed in this paper will consequently get its quality improved by combing the error-detecting system.