Detection Of Masses In Digital Mammogram Based On A Correlation Feature Selection Algorithm

Breast cancer is considered a major cause of threatening the health of women.People begin to make more and more attention for the early stage diagnosis of cancersand digital mammography is the most capable assistant in helping the expert todiagnosis of the cancer. However, some lesions sometimes are hard to be identified byradiologists due to the low quality images and people’s faults. In the study of thebreast’s cancer, people find that with the aid of CAD, not only the subtle lesions canbe found, but also the accuracy rate of the diagnosis can be greatly raised. As a result,the patients will take more time for precautions and treatment and finally make theincidence and death rate of this disease drop considerably.Recently, Digital mammogram has become one of the most effective techniquesfor early breast cancer detection.The aim of this study is to develop acorrelation-based feature selection algorithm to effectively detect the masses in digitalmammogram. Firstly, the regions of interest (ROIs) in the mammogram aresegmented by a topographic representation method. Subsequently, the textural,intensity and shape features are extracted from the ROIs. In order to improve thedetection efficiency and final results, then the correlation-based feature selectionmethod is used in this paper to optimize the feature set and the cost-sensitive learningmethods are introduced here. Finally, we use the optimal set of features to train theBP neural network and classify the ROIs as either masses or non-masses. Theexperimental results show that our proposed methods can more effiectively improvethe detection results than some other popular algorithms in this field.