Automatic Mammogram image Breast Abnormality Detection and localization based on the combination of k-means and Genetic algorithms methods

Abstract

Currently, most oncologists recommend mammography as an effective medical imaging modality for the screening of breast cancer. The aim is to reduce the mortality rate due to this disease. However, several imperfections contribute to the misclassification of breast lesions. Therefore, computer-aided diagnosis (CAD) systems are tools that allow the radiologists a second opinion to improve the diagnosis accuracy. In this paper, an accurate fully automatic method for breast abnormality extraction is presented. The extracted region of interest(ROI) can be subjectively classified by radiologists or used to automatically extract the features that allow for automatic classification. The proposed method consists of two steps: the preprocessing step for delimiting the ROI by removing all the artifacts including the pectoral muscle, followed by the contrast enhancement; the second step is devoted to the extraction of the suspicious area by segmenting the ROI using the k-means algorithm to avoid any initialization. Both contrast enhancement and ROI segmentation are performed by using a genetic algorithm (GA) to achieve good results. The simulation results show the accuracy of the proposed method by comparing the center coordinates and the radius enclosing the abnormality found with those provided with the database.