An analysis on breast tissue characterization in combined transform domain using nearest neighbor classifiers

Abstract

Mammography is a well established imaging technique for showing tissue abnormalities of breast and has been proven to reduce death rate due to breast cancer in screened populations of women. The proposed method classifies the breast tissues according to severeness of abnormality (benign or malign) using combined transform domain features. The discrete wavelet transform (DWT) features are merged with discrete cosine transform (DCT) features. The method is tested on 212 mammogram images from the MIAS database. The cascaded transform domain proves to be a promising tool for robust classification. It yields 94.39% of accuracy in classification of normal and benign samples, 90.65% of accuracy in classification of normal and malign samples and 78.50 % of accuracy in classification of benign and malign samples. Classification is done with a combination of nearest neighbor (NN) classifiers; kNN, class based NN and density based NN.