Large Scale Semantic Annotation of Radiology Reports

Abstract

The development and testing of automated information extraction (IE) systems depends on semantically annotated free text. This presentation reports on the results of a large scale annotation project of a radiology corpus, the Roentgen corpus, consisting of 594,000 deidentified radiology reports with 36 million words, and 4.3 million sentences supplied by Indiana University. The presentation highlights the (1) sentence-based approach in defining propositions annotating the corpus, (2) as well as the annotation framework that is incrementally built and refined in order to facilitate the process of annotation.