Automatic Breast Tumor Classification Using a Level Set Method and Feature Extraction in Mammography

Breast cancer is one of the leading factors of cancer-related deaths among women, therefore designing Computer Aided Diagnosis (CADx) systems to detect malignant and benign tumors of breast masses is extensively essential. Using a segmentation method and subsequently a proper feature extraction is crucial to obtain an appropriate performance in CADx system. In this paper, the Mammography Imaging Analysis Society (MIAS) data is used in order to detect whether breast masses are malignant or benign. A method is based on level set and with the purpose of segmenting the region of the tumor in mammography for the first time and in following, four additional methods were introduced. Including wavelet transform, Gabor wavelet transform Zernike moments and Gray-Level Co-occurrence Matrix (GLCM) to extract features and each one leads to the extraction of a group of the segmented tumor features. Proper features are selected using P value. Consequently, in order to investigate the efficiency of selected features, each group of features are used within one Multilayer Perceptron (MLP). In this paper, the results focus on the appropriate efficiency of proposed segmentation and features extraction methods. Among these considered features, the features related to GLCM are among the best results with the accuracy of 93.37 and sensitivity of 94.18.