The term ontology was originally used in a branch of philosophy (metaphysics). In this discipline, entities are ordered according to their similarities and differences in categories (also known as concepts) with the resulting conceptual arrangements named ontologies. In philosophy categorisations are a prized end point. In Computer Science they are used to semantically enrich data, forming what are referred to as knowledge bases as distinct from databases. A key difference is that knowledge bases can be interpreted independently of how they have been created and used in the past.
The REAMS collection has proven a valuable resource for teaching and diagnosis related to skeletal dysplasias. It captures radiological findings and relates them to a large reference set of images indexed to patients diagnosed with skeletal dysplasias. Over the past two years we have been working on a new representation of the database behind REAMS; one which semantically enhances the radiological findings in REAMS. This has involved developing an ontology of radiology rich enough to classify, deconstruct and interrelate its findings, and mapping the REAMS data against this ontology to form a radiological knowledge base – dREAMS.
The dREAMS ontology has been designed with a view to supporting a level of reasoning, in order to better support analysis. The knowledge base comprises a set of radiographic images with associated patient and radiological phenotype information. The radiological information is represented as a set of findings for each image. A finding is defined as a qualified feature, which in turn comprises an abnormality and its localisation: for example severe [qualifier] bowing [abnormality] of the femur [localisation]. The ontology incorporates an anatomical model for expressing localisation that also includes feature classification hierarchies; and a classification of abnormalities that includes common terms from radiological vocabulary. The knowledge base also links to related data sources such as FMA (anatomy), HPO (abnormalities) and OMIM (diseases and genes).
The model is designed to capture domain knowledge beyond that in the original REAMS database. For example, it captures a semantics rich enough to allow delayed maturation of the epiphyses of the elbow to be interpreted as having localisation [distal epiphysis humerus, proximal epiphysis radius, proximal epiphysis ulna]; but flexion deformity [elbow] to be interpreted as refering to an abnormality of the elbow joint.