The Research On The Extraction Of Temporal Information From Chinese Medical Narrative Records

With the development of clinical informatics and the gradual adoption of clinical information system, such as Electronic Health Records and Clinical Decision Support Systems, it is becoming a hot area in the researches of medical informatics for text mining and information extraction oriented to Chinese medical narrative records. For the high frequency of temporal information in medical records, it is significant to extracting the temporal information and temporal relationship in medical records for realizing medical informatics structuring and promoting the utilization of clinical decision support and medical information mining. Consequently, this paper focuses on three key issues, temporal expression identification, temporal information normalization, temporal relationship extraction oriented to Chinese medical narrative records.Temporal expression identification is one of the key technologies for temporal semantic annotation, and the recognizing results have directly effects to the further usage of temporal information. Consequently, it's necessary to study the methods of recognizing temporal expressions oriented to Chinese narrative medical records, which is an inevitable stage to the usage of temporal information in Chinese narrative medical records. Firstly, this study made a statistic and classification to 147 shares of practical medical records covering more than 30 section offices, and made an analysis to the pattern of time expressions in Chinese medical records. Based on the analytic results, this study proposed a method which applied regular expressions to recognize temporal expressions in Chinese medical records, and applied the proximity principle to recognize composite temporal expressions. The recognizing results showed that the method we proposed could cover most temporal expressions in the corpus referred above, and this study made an important significance and important reference value for further use of temporal information in Chinese narrative medical records.Temporal information normalization is the basis in the research of temporal reasoning. With the results of temporal expression identification, this paper summarized the rules to choose the reference time, based on which, we can calculate the temporal information in medical records described in the calendar. At last, the results were represented as a normalized form, which referred to international standards on temporal annotation. The achievement of temporal information normalization provided standardized data for the follow-up utilization of temporal information. This method can covered more than 96% temporal information in Chinese medical narrative records, and the accuracy can reach 91%.Establishing the relationship between medical problems and temporal information can drive the automatic utilization of temporal information in medical records. For this reason, this study proposes a solution to automatically extract temporal attributes of medical problems from Chinese narrative medical records based on Conditional Random Fields (CRF). In this solution, the medical records were firstly semantically annotated with medical problem and the results of temporal information normalization to fulfill the CRF training task. In the labeled training dataset the temporal relationship was tagged based on medical problem oriented mode, that is to say only interested medical problem's temporal attributes were tagged. A deeply analysis of the impacts of various feature templates of CRF on temporal relationship extraction was taken in such solution framework. A multiple cross-validation method was used to evaluate different CRF learning templates in the corpus which comprised 63 practical narrative medical records, the general principle of template design was proposed. The results showed the accuracy of temporal relationship extraction using the choice template file could reach 86.94%.This paper made a combination of the above three parts, and established initially the approaches to the automatic extraction of temporal information and temporal relationship in Chinese medical narrative records, having achieved good results.