Normal mammogram analysis

Yajie Sun, Purdue University

Abstract

Breast cancer is the leading cause of cancer death among women. Screening mammography is the only method currently available for the reliable detection of early and potentially curable breast cancer. Research indicates that the mortality rate could decrease by 30% if women age 50 and older have regular mammograms. The detection rate can be increased 5–15% by providing the radiologist with results from a computer-aided diagnosis (CAD) system acting as a “second opinion.” However, among screening mammograms routinely interpreted by radiologists, very few (approximately 0.5%) cases actually have breast cancer. It would be beneficial if an accurate CAD system existed to identify normal mammograms and thus allowing the radiologist to focus on suspicious cases. This strategy could reduce the radiologist's workload and improve screening performance. In this dissertation, we propose a new full-field mammogram analysis method focusing on characterizing and identifying normal mammograms. A mammogram is analyzed region by region and is classified as normal or abnormal. We present methods for extracting features that can be used to distinguish normal and abnormal regions of a mammogram. A set of 86 features from four different types of characterization is extracted from each region. We implement a method to select a nearly optimal feature subset for the classification. We propose a unique multi-stage cascading classification method to boost the classification performance. The classifier performs better than a single classifier in that it significantly reduces the false positive rate, the misclassification rate of normal mammograms as abnormal. We have tested this technique on a set of ground-truth full-field mammograms. The results are comparable to human readers. This approach is independent of the type of abnormalities and may complement computer-aided detection based on the recognition of specific types of abnormal structures.

Degree

Ph.D.

Advisors

Delp, Purdue University.

Subject Area

Electrical engineering|Biomedical engineering|Medical imaging

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS