Despite the large volume of terms in RadLex, the lexicon of radiology terms meant to standardize radiologists’ language, terms are still missing and complexities in definitions are apparent, according to a study published in the November issue of Academic Radiology.

Ryan Woods, MD, MPH, of Baltimore’s John Hopkins University School of Medicine, and colleagues chose to investigate the degree to which RadLex contains terms commonly used in clinical practice. The researchers estimated the completeness of RadLex in the realm of chest radiography, characterizing absent terms by collecting a random sample of 100 chest radiograph reports from three clinical radiologists in October 2010.

Woods and colleagues examined the findings and impressions sections of these reports, parsing them into “objects,” which they defined as any discrete physical object, body part, observation, descriptive modifier, diagnosis, or procedure. Objects were then compared to RadLex by entering them into the RadLex term browser. Positive matches was constituted by finding either an exact match, a synonym, or a sufficiently similar meaning. These matches were assigned to one of the following RadLex categories: anatomic object, physiological condition, physical object, imaging observation, procedure, or other. If an object didn’t have any matches, it was placed in the most appropriate category.

Descriptive statistics were calculated for the identified objects including the number of unique objects, the number of unique objects per report, and the frequency of total and unique objects for each RadLex category.

Of the 1,697 object identified, 339 were unique and 211 were present in RadLex, resulting in an overall match rate of 62 percent between the objects and the RadLex term search results. Differences across RadLex categories were statistically significant.

The lexicon was most complete for anatomic objects and physiological conditions, and least complete for imaging observations and procedures. Challenges for RadLex included multiple definitions for a single term, underrepresentation of common physiological conditions, an absence of procedures, and RadLex definitions not falling in line with expected definitions.

An increase in RadLex’s degree of completeness and a refinement of its definitions is feasible, most likely through painstaking yet effective manual methods.

“It’s a double-edged sword,” Woods told Health Imaging. “It would be incredibly useful to have this common language, but the system needs to be user-friendly and comprehensive to do so. If we followed standardized reporting, it could open the door to research and data mining. This is a powerful way to use data that is inherent in what we generate.”