Classification of breast tissue density in digital mammograms

Computer Aided Diagnosis systems (CAD) are the recent trend in digital mammography which helps radiologist in detection of breast cancer. Recent studies are showing that mammography density is strongly associated with breast cancer and it is one of the important risk indicators of this disease. Automatic detection of breast lesion is depends on effective pre-processing and breast tissue type classification. Hence this paper proposes a new method for mammogram image pre-processing and tissue type classification. In pre-processing automatic removal of noises, labels and artifacts is performed by morphological operations followed by connected component labelling and pectoral muscle is eliminated by polynomial curve fitting method. Statistical features like Mean, Standard deviation, Smoothness, Skewness, Uniformity, Kurtosis, Histogram and correlation are the important texture features of the breast tissue are extracted and fed into Support Vector Machine (SVM) for classification. Our method is experimented on Mini-MIAS database (Mammographic Image Analysis Society, London, U.K.) for classification of mammogram into any of the three classes namely Fatty, Dense and Glandular and it yields good result when compared with existing techniques.