Abstract
Abstract: :
Purpose:Ophthalmic images have an implicit semantics that must be made explicit for efficient organization, storage and retrieval. We have created an ontology (a formal specification of a domain) using XML and RDF that describes ophthalmic image attributes and their relationships. We have further instantiated a database according to this data model, allowing accurate and machine-readable characterization of image semantics. Methods:The ontology was created as an RDF schema using the knowledge-modeling tool Protégé-2000 (http://smi-web.stanford.edu/projects/protege). RDF is an XML application that encodes knowledge as triples consisting of a resource (subject) with a property type (predicate) and its value (object). An object-oriented model was followed that defines classes in a hierarchical tree, with child classes inheriting properties from parent classes. Instances represent real-world examples of classes. A database of fundus images was annotated according to the schema and a set of query and inferencing tools were written in the Java programming language. Results:The ontology includes the classes (image types) Angio_Image (flourescein and ICG), Laser_Image (OCT), MRI_Image (MRI), Radiog_Image (CT and X-Ray), Ultrasound_Image (A scan, B scan, and biomicroscopic) and VisibleLight_Image. This latter category was further divided into Mapped_Image (e.g. visual field or corneal topography) and NonMapped_Image (e.g., external or fundus photos). A wide variety of class attributes were modeled as well as constraints on attribute values. The ontology was used to populate an ophthalmic image database comprising fundus images from 398 diabetic patients. The ontology further allowed for Boolean queries of the database and inferencing based on the structured metadata expressed in RDF. Conclusion:We have created an ontology for ophthalmic images and instantiated an image database according to its specifications. The resulting schema defines resources and properties of the most common image types. It creates a structured data model that allows for automatic severity assessment and complex querying of image metadata. This allows efficient computing over image sets and should also foster data mining and machine learning applications over these sets. A wide variety of medical image types may benefit from metadata-based ontologies to organize images for research, teaching, and clinical care.
Keywords: 356 clinical (human) or epidemiologic studies: systems/equipment/techniques • 350 clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • 355 clinical (human) or epidemiologic studies: risk factor assessment