NegBio: a high-performance tool for negation and uncertainty detection\n in radiology reports

Abstract

Negative and uncertain medical findings are frequent in radiology reports,\nbut discriminating them from positive findings remains challenging for\ninformation extraction. Here, we propose a new algorithm, NegBio, to detect\nnegative and uncertain findings in radiology reports. Unlike previous\nrule-based methods, NegBio utilizes patterns on universal dependencies to\nidentify the scope of triggers that are indicative of negation or uncertainty.\nWe evaluated NegBio on four datasets, including two public benchmarking corpora\nof radiology reports, a new radiology corpus that we annotated for this work,\nand a public corpus of general clinical texts. Evaluation on these datasets\ndemonstrates that NegBio is highly accurate for detecting negative and\nuncertain findings and compares favorably to a widely-used state-of-the-art\nsystem NegEx (an average of 9.5% improvement in precision and 5.1% in\nF1-score).\n