Use of imaging technologies within healthcare delivery organizations has grown dramatically over the last decade, providing previously unavailable diagnostic and screening capabilities. At the same time, growth in the number of reports and images generated contribute to growing challenges to optimally use clinical information while not being overwhelmed by it. In addition to providing reports addressing questions posed by ordering providers, radiologists often identify unexpected incidental findings that may pose a significant health risk to the patient in the short or medium term. Professional societies have assisted in this effort by defining what types of findings justify designation as critical results, specified how these results should be communicated, and provided guidelines for the follow-up of incidental finding communication. However, there is an unmet need to provide an automated system infrastructure to track and identify both critical test results and unexpected findings as these unexpected findings have been shown to “fall through the cracks” alarmingly frequently.
Communication of follow-up recommendations when abnormalities are identified on imaging studies is prone to error. The absence of an automated system to identify and track radiology recommendations is an important barrier to ensuring timely follow-up of patients especially with non-acute incidental findings on imaging examinations. In this webinar, we present a text processing pipeline to automatically identify clinically important recommendation sentences in radiology reports. Our extraction pipeline is based on natural language processing (NLP) and supervised text classification methods. To develop and test the pipeline, we created a corpus of 800K radiology reports from three different institutions. We ran several experiments to measure the impact of different feature types and the data imbalance between positive and negative recommendation sentences.
After participating in this activity, the learner should be better able to:
- Understand the application of NLP in the radiology domain
- Discuss the current recommendations in radiology reports
Meliha Yetisgen, PhD
University of Washington
Meliha Yetisgen is an Associate Professor in the Department of Biomedical Informatics and Medical Education and Adjunct Associate Professor in the Department of Linguistics at the University of Washington (UW). She leads the UW-BioNLP research group. Before joining UW, she worked as a researcher in industry and served on the advisory boards of several text mining startups. Her current research interests include statistical natural language processing, machine learning, and biomedical/clinical text mining. Dr. Yetisgen received her BS degree in Computer Engineering from Bilkent University (Ankara, Turkey) and MS degree in Computer Engineering from Middle East Technical University (Ankara, Turkey). She received her PhD from University of Washington with a thesis on automated hypothesis generation from biomedical literature.