David A. Koff MD, Presenter: Nothing to Disclose

Stuart A Doherty BA, Abstract Co-Author: Nothing to Disclose

Content-Based Image Retrieval (CBIR) technology introduces a new paradigm in the way we think about navigating and searching large image databases. With the rapid growth in size and scope of medical diagnostic image databases, new tools are needed to help experts in the field efficiently search and intelligently harness the information within these repositories.

Operating on the pixel values and highlighted objects within the image, CBIR technology strives to compress the visual properties of an image into a concise representation or signature. An image signature captures fundamental image properties such as color, texture, shape, and location as they relate to specific types of pathology. To increase the speed of search, and decrease the space required to index a repository, special attention must be paid to the design of the index. In this research the image repository is composed of CT scans and MR cases of cerebral lesions including tumors, hemorrhages, and abscesses. A prototype implementation has been developed using C++ and Matlab on the Linux operating system.

This tool presents case-relevant images exhibiting a high degree of visual similarity with respect to the pathology bearing regions (PBRs). Visually similar example images, and their associated diagnosis, provide radiologists conducting an assessment content with which they can compare and contrast.

CBIR technology presents an effective tool for aiding in diagnosis. Additionally, the CBIR framework provides for a richer experience when navigating a database. Radiology residents are not restricted anymore to text-based search , but can enhance their diagnostic skills in easily accessing visually comparable images.

Koff, D, Doherty, S, A New Extensible Framework for Research In Content-based Medical Image Retrieval.  Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL. http://archive.rsna.org/2005/4415039.html