This issue of Academic Radiology contains a study by Woods and Eng describing their experience using chest radiography reports generated in usual clinical practice to estimate the completeness of RadLex ( http://rsna.org/RadLex.aspx ). RadLex was developed by the Radiological Society of North America to meet the terminology challenges of radiology by providing a uniform source of terms and concepts used by and for radiologists. The intent of RadLex is to reduce variation and improve clarity in radiology reports and image annotations as well as to provide a standardized means of indexing radiological materials in a variety of settings. These standardized terms can then be used for retrieving information from a variety of imaging information sources, including imaging research databases, educational materials, and clinical imaging reports. Using clinical imaging reports as the data source, this study is an important critique of the ability of RadLex to meet its stated goals.
The study looked at the match rate for “objects” in routine chest radiography reports. In the analysis, there are two critical RadLex gaps that the authors identified that reduce matching frequency. The first is in relation to the terms that describe patient procedures, for example, “coronary artery bypass” or “aortic valve replacement.” The initial impetus for RadLex was identified through a process of report content parsing and evaluation that was similar to the current study. Langlotz and Caldwell mapped radiology reports objects against the largest medical lexicons, the Unified Medical Language System, and two constituent terminologies, International Classification of Diseases, Ninth Revision, Clinical Modification and SNOMED International. The maximum coverage between radiology terms and content in the standard medical lexicons was 50%. In the current study, the procedure terms that fail to have matches in RadLex are precisely those terms that are readily found and well defined in those clinically oriented lexicons.
RadLex has a tool in its term browser web page ( http://radlex.org ) by which new content, including the phrases that constituted a nonmatch in research studies such as this, might be suggested. The addition of these phrases would improve direct content matching between a “radiology” lexicon and radiology report objects. However, it is critical to question whether expansion of RadLex content better meets the needs of the radiology community and where and how content that is contained in other ontologies should be accessed.
RadLex is intended as a lightweight application ontology that already has forged partnerships with larger reference ontologies, like the Foundational Model of Anatomy in order to improve ontologic content exchange and interoperability . Acknowledging that medical and clinical content is indexed in multiple ontologies and lexicons, the National Centers for Biomedical Computing hosts a website ( http://bioportal.bioontology.org ) of more than 350 ontologies. Searches through this website can identify a multitude of objects within diverse ontologies to which a particular phrase might map.
RadLex content is dynamic, with constant updates, additions, and realignments. During the study period cited in the article, November 2010–November 2011, four versions of RadLex were released. The final version (3.5) used for the study contained 34,435 classes, where a class represents a distinct ID in RadLex and 58,025 instances. All RadLex terms, whether preferred terms or synonyms, are slotted as instances. The instances are assigned the same class of RadLex ID, accounting for the greater number of instances than classes. In the most current working version of RadLex 3.9, there are 42,321 classes and 69,151 instances (B. Collins, RadLex curator, personal communication). Thus, if the study were completed today, the results of the study might be different given that the source for term matching has increased by nearly 20%.
Additionally, by tailoring the object matching criteria to the domains best represented by RadLex, the expected match rate might be higher than that found in the current study. The authors describe the Marwede et al study of radiology reports as one that found a significantly higher match rate, at 84% (range 78–90%), then the current study with an overall match rate of 62% (range 0–77%). The current study took a broad approach to the matching task and included common domains within chest radiograph reports including procedures and imaging procedure attributes. The stated goal in the task of object definition was to be as conceptually diverse as possible while maintaining a high degree of granularity. The diversity of concept classification exposed important gaps in RadLex.
Get Radiology Tree app to read full this article<
Get Radiology Tree app to read full this article<
Get Radiology Tree app to read full this article<
Get Radiology Tree app to read full this article<
References
1. Woods R., Eng J.: Estimating the completeness of RadLex in the chest radiography domain. Acad Radiol 2013; pp. 1329-1333.
2. Langlotz C.P., Caldwell S.A.: The completeness of existing lexicons for representing radiology report information. J Digit Imaging 2002; 15: pp. 201-205.
3. Mejino J.L., Rubin D.L., Brinkley J.F.: FMA-RadLex: an application ontology of radiological anatomy derived from the foundational model of anatomy reference ontology. AMIA Annu Symp Proc 2008; pp. 465-469.
4. Brinkley J.F., Detwiler L.T., Group S.I.: A query integrator and manager for the query web. J Biomed Inform 2012; 45: pp. 975-991.
5. Duftschmid G., Wrba T., Rinner C.: Extraction of standardized archetyped data from electronic health record systems based on the entity-attribute-value model. Int J Med Inform 2010; 79: pp. 585-597.
6. Marwede D., Schulz T., Kahn T.: Indexing thoracic CT reports using a preliminary version of a standardized radiological lexicon (RadLex). J Digit Imaging 2008; 21: pp. 363-370.
7. American College of Radiology: Breast imaging reporting and data system atlas (BI-RADS ® Atlas).4th ed2003.American College of RadiologyReston, Va
8. Hansell D.M., Bankier A.A., MacMahon H., et. al.: Fleischner Society: glossary of terms for thoracic imaging. Radiology 2008; 246: pp. 697-722.
9. Hong Y., Zhang J., Heilbrun M.E., et. al.: Analysis of RadLex coverage and term co-occurrence in radiology reporting templates. J Digit Imaging 2012; 25: pp. 56-62.