The analysis of digital mammograms: Spiculated tumor detection and normal mammogram characterization
Abstract
Mammography is currently the best technique for reliable detection of early, non-palpable, potentially curable breast cancer. In the past decade there has been tremendous interest in computer aided diagnosis (CAD) in digital mammograms. The goal has been to increase diagnostic accuracy as well as the reproducibility of mammo-graphic interpretation. Among breast abnormalities, spiculated masses having a stellate appearance are the most difficult type of tumor to detect. We propose a multiresolution feature analysis and classification approach for the detection of spiculated lesions. In general, it is difficult to determine the size of the neighborhood that should be used to extract local features of spiculated lesions. There is a fundamental difference in our method compared to other approaches, in that we extract and classify features at multiple resolutions, hence overcoming the difficulty of choosing a neighborhood size a priori to capture tumors of varying sizes. Furthermore, the top-down classification requires less computation by starting with the coarsest resolution image and propagating detection results to finer resolutions. On the MIAS database, we achieved 84.2% true positive detection at less than 1 false positive per image and 100% true positive detection at 2.2 false positive per image. The effectiveness of algorithms for detecting cancers can be greatly increased if these algorithms work synergistically with algorithms for recognizing normal mammograms. However, little work has been done on understanding normal mammograms. We propose a novel technique to specifically characterize normal mammograms based on normal tissue identification and removal, which is independent of the types of abnormalities that may exist in the mammogram. Experimental results have shown that this approach also facilitates the classification of abnormalities, since suppressing normal background structures enhances the contrast and obviousness of abnormal structures.
Degree
Ph.D.
Advisors
Delp, Purdue University.
Subject Area
Electrical engineering|Biomedical research|Oncology
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.