A new approach to extract meaningful clinical information from medical notes

摘要

In medical domain, one of the most important documents is the notes that doctors, nurses or other medical practitioners take during patient interview. These notes contain important information about the patient current condition, symptoms, family history, disease, procedures (like x-ray, lab test etc.), medication and so on. These notes are in plain text English language and are presented in an unstructured way. An unstructured text can be defined as text which contains information not in a common structured format. Since these notes contain very useful clinical information, everyone is willing to extract such information. But lack of defined structure, it makes these texts to be interpreted only by humans and not by any computer program. In structured information, we have only the useful information extracted from the text and other unnecessary information is thrown away. This makes the information to be presented in a definite structure which can be stored in database for further processing. In this approach, we are going to build a medical corpus using the training dataset and apply the corpus in order to extract the information like diagnosis, procedure, drug, habit and vitals from the clinical notes and evaluate it based on different accuracy parameters.