Secure and efficient retrieval of multimedia information from archives of large biomedical images for teaching, research and development of diagnostic tools have become a necessity in today's global environment. Most traditional content-based image retrieval (CBIR) systems do not address the challenges involved in high fidelity and efficient retrieval of diverse classes of biomedical imagery through secure transmission over the Internet. We present the development and validation of a source encoding scheme in the wavelet domain with an embedded backward coding of wavelet trees (BCWT) for encoding of multiple image classes archived at the National Library of Medicine. HVSQ/BCWT can be used for fast and secure retrieval of significant volumes of medical diagnostic information over the Internet. Work in automated segmentation of such images using deterministic annealing and Gaussian mixture modeling is also presented.