Integrating content-based visual access methods into a medical case database.

Abstract

In the computer vision domain, content-based access methods to all forms of multimedia data are a hot research topic. A large number of tools have been developed to find documents in multimedia repositories and to manage the (visual) information that has been created, for example by the Internet. Although no general breakthrough has been achieved with respect to searching diverse databases with automatically extracted features, the techniques have gained acceptance in a few well-defined domains such as image agencies (Corbis) and trademark research. In the medical domain, the number of digital images produced and used for teaching, diagnostics and therapy is rising in a similar way as in other domains. Still, there are only a few image retrieval systems that use automatically extracted visual features for content-based access to medical image databases. This article describes the use of an open source image retrieval system (GIFT) that has been adapted for the use with medical images using the image case database CasImage that has been developed and maintained by the University Hospital of Geneva and that is in routine use. The first results show the potential of this technique to retrieve similar cases from a case database. This is an important task for teaching and might also become important for diagnostics using case-based reasoning, for example. For the use as a diagnostic tool, it is foreseen to specialize the visual features for the domain of lung image retrieval, using high resolution computed tomography (HRCT) images.