Computer Aided Analysis of Mammograms for Breast Diseases

Abstract

Breast cancer is the life-threatening disease that invades into surrounding tissues or newlineextends to distant body parts. It is the most prevalent cancer amongst women across newlineworldwide. The regular screening of breast is essential to investigate cancer risk in its newlineearly stage. The early detection of breast cancer is necessary to initiate radiation therapy newlineto stop or control further growth of it. Mammography is considered as a gold standard newlinefor screening of breast, which can detect lump or tumor before actually it is felt. The newlinemammogram images are X-ray images, which are captured using low dose X-ray or very newlinelimited count of X-ray photon. Therefore, mammogram images are poor contrast images, newlineweak edges and it also consists of random fluctuations, so that interpretation of newlinemammogram is one of the difficult tasks for radiologist. The study shows that 10-30% newlinebreast cancers are misinterpreted by radiologist. Therefore, Computer Aided diagnosis newlineand detection (CAD) system act as a second reader for radiologist, which performs newlineautomatic segmentation and detection of breast cancer. The contrast enhancement of newlinemammogram is an important preprocessing stage for CAD system to better visualize newlinecontents of mammogram. In this thesis, a novel image fusion using Local entropy newlinemaximization technique for mammogram contrast enhancement is proposed. The newlineproposed contrast enhancement technique enhances contrast and improves the edge newlinecontents. To overcome the problem of false positives detected in automatic breast tumor newlinesegmentation, this thesis proposes the false positive reduction as a posterior step of CAD newlinesystem. Since, the segmented suspicious region may not be abnormal called as false newlinepositive region, which consumes radiologist time and sometimes may results into newlineunnecessary biopsies. Therefore, in the proposed system, the initial suspicious region is newlinesegmented using self-organizing map (SOM) network, then the detection of false positives newlineand subsequently removal or reduction of false positives are achieved using feature newlineextraction an