Classification Driven Medical Image Retrieval

Abstract

We propose a novel image retrieval framework centered around classification-driven search for a weighted similarity metric for image retrieval. This approach is firmly rooted in Bayes decision theory. Given a well-defined image set, we argue that image classification and image retrieval share fundamentally the same goal. Thus, the distance metric defining a classifier that performs well on the image set should also generate good results when used as the similarity metric for image retrieval. In this paper we report our methodology and initial results on neuroradiological image retrieval, where the approximate bilateral symmetry of normal human brains is exploited.