Revisiting the Feature and Content Gap for Landmark- Based and Image-to-Image Retrieval in Medical CBIR

Medical image content-based retrieval entails several possible scenarios. One scenario relates to retrieving based on image landmarks. In this scenario, quantitative image primitives are extracted from the image content, in an extensive pre-processing phase, following which these quantities serve as metadata in the archive, for any future search. A second scenario is one in which image-to-image matching is desired. In this scenario, the query input is an image or part of an image and the search is conducted by a comparison on the image level. In this paper we review both retrieval scenarios via example systems developed in recent years in our lab. An example for image landmark retrieval for cervix cancer research is described based on a joint collaboration with National Cancer Institute (NCI) and the National Library of Medicine (NLM) at NIH. The goal of the system is to facilitate training and research via a large archive of uterine cervix images DEFINING THE IMAGE RETRIEVAL TASK Medical content-based image retrieval (CBIR) deals with retrieving visual information from medical images. For an extended overview of the role of CBIR within medical information retrieval and health systems such as Picture Archiving and Communications Systems (PACS), see Smeulders, Worring, Santini, Gupta, and Jain, (2000); Müller, Michoux, Bandon, and Geissbühler, (2004); Lehmann, Antani, and Long, (2004). In this paper