| Canadian women have a one in nine chance of developing breast cancer during their lifetime; early detection is the best way to increase survival following diagnosis. The disease cannot be detected and treated with a single technique or methodology because manifestation of the disease is unique to each woman. A combination of information, techniques, and technologies from different fields of study is needed to address the problem of detection, diagnosis, and treatment of breast cancer. The design of the comprehensive University of Calgary indexed atlas of mammograms (U of C atlas) incorporating retrospective patient and image information, image processing techniques, advanced query and retrieval capabilities, and enabling technologies is presented in this thesis. The atlas is intended to assist radiologists with effective and efficient diagnostic support capabilities. The U of C atlas design is based on medical reasoning, the workflow of the radiologist, standard BI-RADS™ definitions and nomenclature, and computer-aided image analysis.; Image processing techniques were used to extract parameters related to the shape and texture of breast masses. The parameters provide quantitative assessment of mammographic features that could be used to represent the masses in the index of the atlas. Classification based on parameters extracted via shape analysis indicated that the fractional concavity values were able to discriminate well between malignant and benign masses with a specificity of 94% and sensitivity of 85%, thus supporting their use in the indexing scheme of the U of C atlas.; A content-based retrieval strategy using the Euclidean distance between vectors of shape feature values was implemented in Matlab®. Given a query sample of a mass, similar masses are retrieved and presented to the user in the order of the most similar to the least similar mass. Content-based retrieval efficiency was evaluated based on the number of correctly classified masses that were retrieved from the atlas. Precision values ranging from 86% to 89% were realized with various combinations of shape factors.; Advances in database design to incorporate information fusion, and advanced computational models such as parallel computing, grid computing, mobile software agents, and web-based applications are presented as future implementation options for computer-aided diagnosis of breast cancer. |
