Identifying image structures for content-based retrieval of digitized spine x-rays

We present ongoing work for the computer-assisted indexing of biomedical images at the Lister Hill National Center for Biomedical Communications, a research and development division of the National Library of Medicine (NLM). For any class of biomedical images, a problem confronting the researcher in image indexing is developing robust algorithms for localizing and identifying anatomy relevant for that image class and relevant to the indexing goals. This problem is particularly acute in the case of digitized spine x-rays, due to the projective nature of the data, which results in overlapping boundaries with possibly ambiguous interpretations; the highly irregular shapes of the vertebral bodies, sometimes additionally distorted by pathology; and possible occlusions of the vertebral anatomy due to subject positioning. We present algorithms that we have developed for the localization and identification of vertebral structure and show how these algorithms fit into the family of algorithms that we continue to develop for our general indexing problem. We also review the indexing goals for this particular collection of digitized spine x-rays and discuss the use of the indexed images in a content-based image retrieval system.