Hybrid feature based method for early detection and classification of mammographic density in breast cancer diagnosis

Abstract

Breast cancer is a common disease with unknown reason. So, early detection or prediction is better way to fight against this disease. A method can be used to detect mass in mammograms and predict breast cancer in its early stage. This early detection of breast cancer will play an important role to decrease the mortality rate due to this disease. The method, in general, consists of some sections like: image pre-processing, pectoral muscle segmentation, suspected area identification, feature extraction and classification. The pectoral muscle extraction and removal from the mammogram is a prerequisite to gain more accurate mammographic density detection and classification. Here, the pectoral muscle segmentation has been performed by region grow method. Then the mammogram without pectoral muscle has been divided into sixteen-by-sixteen blocks and by taking the Gray Level Co-occurrence matrix (GLCM) of the blocks, the region of interest (ROI) containing possible mammographic density has been extracted for further analysis. Intensity-based or area-based or hybrid of both the features based extracted from the segmented region of interest (ROI) of the original mammogram have been used for classification. By using Support Vector Machine (SVM) method or two-layer Neural Network (NN) method, the ROI of the mammographic density was classified as the normal, benign or malignant. Simulations have been performed on Mammographic Image Analysis Society (MIAS) mini database and effectiveness of the method was shown by different performance. Correction rate has been found 96.77% for SVM and 95% for Feed Forward Neural Network. The proposed method of mammogram screening for early detection of breast cancer can open a new era to biomedical field and make it easier to decrease mortality rate caused by breast cancer.