RSNA 2017: Imaging Informatics Preview

Ontology enhances mining of radiology reports

By Erik L. Ridley, staff writer

October 30, 2017 --

Tuesday, November 28 | 10:50 a.m.-11:00 a.m. | SSG07-03 | Room N230B
In this scientific presentation, researchers will show how ontology-driven mining of free-text radiology reports may help improve decision-making and find new associations between imaging manifestations and diseases.

Interest is increasing in phenomics, or the ability to extract information from electronic health records to link the manifestations of disease with genes and gene expression, said presenter Dr. Charles Kahn Jr. of the University of Pennsylvania. In their study, the researchers sought to explore the ability to extract information about associations between diseases and imaging findings from conventional free-text radiology reports.

They applied the Radiology Gamuts Ontology of diseases and imaging findings, which expresses the prevalence and co-occurrence of 13,964 diseases, interventions, and imaging findings in a sample of 1.2 million patients. Using terms and synonyms from the ontology, the researchers developed an automated system that could calculate conditional probabilities for two examples of differential diagnosis: diffuse bowel wall thickening and splenomegaly.

"We have demonstrated the ability to extract quantitative information from conventional [free-text] radiology reports, which can be used to improve diagnostic decision-making in radiology and potentially to identify new associations between diseases and their imaging manifestations," Kahn told

How did they do it, and what else did they find? Get all of the details by sitting in on this Tuesday morning presentation.