A shape-based approach for image retrieval in healthcare information infrastructure

Loading...
Thumbnail Image
Date
2008
Journal Title
Journal ISSN
Volume Title
Publisher
University of Zululand
Abstract
This study investigated some models and techniques that would help build shape-based image retrieval with an improved accuracy. As an initial step, a modular prototype system, called BrainSearch was implemented and used to demonstrate the utility of our algorithms and techniques on brain Magnetic Resonance Imaging (MR1) characterization and their suitability for image retrieval. The system supports retrieval based on shape similarity, a single keyword image annotation and five brain MRl classes. The BrainSearch system was realized to make it easy to test retrieval performance and to expedite further algorithm investigation. This was made possible by the implementation of region-based Local Binary Fitting (LBF) active contour, Density histogram of Feature Points (DFP) shape representation and k-Nearest Neighbor (k-NN) classifier. Then we performed a series of experiments to evaluate the performance of BrainSearch utilizing different retrieval techniques. Results generally showed that (a) region-based DFP shape representation is better than edge-based DFP shape representation, whether pre-classification of images is used or not, (b) retrieval technique that uses pre-classification of images gives better results than retrieval technique that uses non-classification of images, no matter the DFP shape representation used, (c) the pre-classification of images cannot improve edge-based DFP shape representation better than when region based DFP alone is used, and (d) the pre-classification of images as well as factors, like shape representations and similarity measures, improve retrieval performance of BrainSearch system. Overall, the hybrid combination of LBF active contour, DFP shape representation and k-NN classifier is promising for the retrieval ofbrain MRI.
Description
Submitted to the Faculty of Science in the Department of Computer Science at the University of Zululand, 2008.
Keywords
Image retrieval, Shape-based image retrieval
Citation
Collections