IRMA Code II: Unique Annotation of Medical Images for Access and Retrieval

Content-based image retrieval (CBIR) provides novel options to access large repositories of medical images, in particular for storing, querying and reporting. This requires a revisit of nomenclatures for image classification such as DICOM, SNOMED, and RadLex. For instance, DICOM defines only about 20 concept terms for body regions, which partly overlap. This is insufficient to access the visual image characteristics. In 2002, the Image Retrieval in Medical Applications (IRMA) project proposed a mono-hierarchic, multi-axial coding scheme called IRMA Code. It was used in the Cross Language Evaluation Forum (ImageCLEF) annotation tasks. Ten years of experience have discovered several weak points. In this paper, we propose eight axes of three levels in hierarchy for (A) anatomy, (B) biological system, (C) configuration, (D) direction, (E) equipment, (F) finding, (G) generation, and (H) human maneuver as well as additional flags for age class, body side, contrast agent, ethnicity, finding certainty, gender, quality, and scanned film, which are captured in form of another axis (I). Using a tag-based notation IRMA Code II supports multiple selection coding within one axis, which is required for the new main categories.