Extracting clinical information from free-text of pathology and operation notes via Chinese natural language processing

Many of surgical records containing the clinical information are in electronic forms, but a lot of them are still in free-text format in China. In this paper, we have an attempt to extract information with the Nature Language Processing (NLP) approach. The procedure of NLP is made up of three steps. First, given 36 free-text of operation notes, a physician manually annotates the information which he is interested in. Second, we extract the features of the annotated information. Third, several logistic regression models are built. Totally, 14 clinical data are extracted. The NLP tool was tested 364 operation notes. The accuracy of extraction is between 67.3%–96.7%. Our results indicate that the performance of the features we used to build the machine learning is good in extracting useful information from free-text Chinese operation notes for liver cancer. In the future, these features would explored on more broader clinical settings.