Classification Driven Semantic Based Medical Image Indexing and Retrieval

Abstract

The motivation for our work is to develop a truly semantic?ased image retrieval system that can discriminate between images differing only through subtle, domain?pecific cues, which is a characteristic feature of many medical images. We propose a novel image retrieval framework centered around classification driven search for a good similarity metric (image index features) based on the image semantics rather than on appearance. Given a semantically well?efined image set, we argue that image classification and image retrieval share fundamentally the same goal. Thus, the distance metric defining a classifier which performs well on the data should be expected to behave well when used as the similarity metric for semantic image retrieval. In this paper we shall report our methodology and results on 3D grey?evel medical image retrieval.