Study Of Information Extraction Methodology In Free-Text Medical Records

Extracting important medical information automatically from free-text medical records is a necessary foundation to provide needed information for Clinical Decision Support, Data Mining and other clinical infromation systems. The flexibility of nature language is a huge challenge for Medical Language Processsing because it has closed relationships with many other research fields such as domain knowledge, grammar knowledge, computing theory and so on. This thesis proposes a new solution to extract information from Chinese free-text medical records based on valuable experience and good results of English medical language processing. This solution has both adaptive advantage from machine learning and high precision from syntactic-semantic parsing, moreover, it is implementated and verified in free-text family medical records. In addition, this thesis proposes an applicable information format of medical records according to all kinds of available medical information standards and common information formats. The success of information extraction from free-text medical records will overcome the information acquisition bottleneck of Clinical Decision Support, Clinical Path Management and other cutting-edge medical information fields, furthermore, it enhances our country's core competiveness of medical information technology and has significant social and economic benefit.This thesis completes the following research works:Firstly, this thesis reviews the new development of medical language processing and its application in clinical environment, compares the similarity and difference between English and Chinese medical language processing and then sums up valuable experience of English medical language which can be used in Chinese medical language processing.Secondly, this thesis uses statistical machine learning method to automatically recognize medical problems in free-text medical records due to the small-scale of Chinese medical terminology database and knowledge base.Thirdly, this thesis uses pattern matching method to extract relationships between important information because it lacks Chinese medical language syntactic-semantic corpus and sentences in free-text family medical records are relatively simple.Finally, the whole solution is completed and verified in free-text family medical records.