Mammogram Images Detection Using Support Vector Machines

Abstract

Breast cancer begins when an abnormal growth of cells takes place in the breast. We formulated a procedure which explains identification of cancer cells in breast cancer X-ray images. This study is useful for doctors to discover abnormal tissues in given set of X-ray images. The initial phase of this procedure intended to enhance the mammogram image sequence. Initial phase is data cleaning phase in which noise is removed and emphasizing the inner structure of the mammogram image. In the second phase CNNs are used to segment the regions which consists of cancer cells. These regions may have various shapes like circular density, eccentricity, density, circularity and circular disproportion. Shape descriptors are used to asses the shapes of the regions of interest. Textures are analyzed with the help of geostatistic functions like Geary's index and Moran's index. SVMs are used to categorize the brain image into two regions such as non-masses and masses, with 0.3 false negative value per mammogram image, 0.86 false positive value per mammogram image, sensitivity 79% and ROC value 90%.