Biomedical image retrieval with self organizing map and relative concept distance normalization(RCDN)

Abstract

Content based medical image retrieval is one of the most intriguing research area of modern image retrieval and biomedical image processing. In the biomedical domain the volume of information is rapidly increasing. For managing this huge amount of volume of information content based medical image retrieval techniques are indispensable. Medical researchers, physiotherapists, dentists and clinicians vividly use online databases to access the relevant biomedical bibliographic citations based on keyword search on different elds such as the main text, author, and date. However, a picture is worth a thousand words, and even more so in medical domain as images of diverse modalities constitute an important source of anatomical and functional information for the diagnosis of diseases, research, education. Researchers have proposed di erent methods to extract the meaningful information from images as well as retrieve images from query image. Though text based image retrieval is a well traversed approach, the e cient way of content based medical image retrieval is always preferred for bridging semantic gaps. We propose a novel approach for medical image retrieval where we represent the image with concepts which refers to distinguishable visual texture and shape definitions representing the visualness of an image. For concept learning we have used self-clustered weighted entropy based concept feature space trained by self-organizing map. We also proposed a new ranking procedure which is based on normalization of histogram of concepts. The hypothesis that such approaches would improve biomedical image retrieval is validated through experiments on ImageCLEFmed2009 IRMA data set, which is collected with o cial permission of IRMA.