Research On The Pre-processing And Detection For The Breast Cancer Automatic Diagnosis System

Breast cancer is the second leading cause of cancer death in women, exceeded only by lung cancer. The mammography has the characteristics of high resolution and large pixel depth, and with these reasons the normal breast cancer detection algorithms will encounter the speed bottleneck. In this paper, we focus on mitigating such a speed bottleneck with following works: fast feature extraction and classification based on a new ROI identifying strategy; a cascade realization of screening true negative regions; design of multi-screening algorithm based on multi-scales with improvement of the ability of screening background.The first work of this paper is giving a new opposite strategy of finding the ROIs. Different from traditional direct methods, the new algorithm finds the normal regions positively containing no MCCs as many as possible, and screens these regions out to identify the ROIs indirectly. This way could avoid computing the complex characteristics of MCCs, and speed up the screening process with the reason that characteristics of normal regions usually are simpler than that of MCCs.Based on the above the first work, the second work of this paper is to further develop a screening algorithm of combing two successive classifiers in a cascade structure which dramatically increases the speed of the screening process. The first simple classifier using threshold method is responsible for fast screening out most of TN regions, and while the second with relatively greatly powerful cost-sensitive support vector classifier (CSSVC) for refining the rest regions.Thanks to the microcalcified spots with various sizes, the third major work is giving a multi-screening algorithm based on multi-scales from low to high scales. The low scale which reflects global information has a insensitiveness to noise, but easily miss small MCC regions. While the high scale which reflects local information is sensitive to small MCC regions and noise, especially the partitioned blocks with the sizes equal to the average diameter of microcalcification. Taking advantage of all the...