Extracting Targets and Attributes of Medical Findings from Radiology Reports in a mixture of Languages

Abstract

This paper introduces machine learning methods for extracting targets and attributes and identifying associations among them from radiology reports written in Korean and English. In the target extraction task, conditional random fields are utilized with language and domain specific features. In the task of finding an association between a target and an attribute, a simple method of generating negative examples from positive examples is introduced and experimented with three different statistical classifiers.