Structurization Of Digestive Endoscopy Report Based On Natural Language Processing

The patient-record is the main carrier of medical information in the process ofhospital management. It is not only the first-hand data of the medical treatment andscientific research but also the evidence of the integrative evaluation for medicalquality, technology and management level. If you want to obtain documents withsome definite condition out of a large number of medical documents and to performanalyzing and summarizing on it, for example, to investigate the rules such as therange of the age and zone of the person infected by certain disease, the relationshipbetween the symptoms and diseases, analysis of the text report is necessary. Medicalrecord is usually inputted with natural language freely by doctors. And this methodleads to massive narrative report. The hospital is a special field and the dailyinformation volume is big. It is very difficult to obtain, inquire and analyze relativeinformation from a large number of free-text medical documents. On the other hand, massive non-structured and non-standardized text information abandons theinformation share and statistics between hospitals. So it is significant to study thestructurization of medical document based on natural language processing.At present, the EMR researchers develop many methods to realize thestructurization of medical documents. The structured entry is a popular method, but itcan not express all semantic information that can be expressed by natural language.This dissertation researches the structurization method for free-text medical recordswith the example of the digestive endoscopy reports. The contents are listed below:Analyze and compare the existing automatic participle tools, select andimplement ICTCLAS as primary tool to divide words in free-text reports.Adjust above primary participle result. Using the special dictionary, distinguish the MST standard terminology, as well as those can betransformed to MST terminology.Analysis the structural character and connotative relationship of MST. Basedon the UMLS Metathesaurus and Semantic Network, establish the MSTsemantics network knowledge library.Analyze the participle result text of gastroscopy report, and then transformthem to structural reports satisfied MST standard, whose accuracy rate is92.3%.It is proved that our method is easy and effective and it provides an experimentalplan for the structural and standard electronic gastroscopy record. Furthermore, it can be applied in other parts of electronic records to realize their structualization andstandardization.