Feature selection for automatic breast density classification

Mammography is probably the best method for early detection of abnormalities in the breast tissue. Higher breast tissue densities significantly reduce the overall detection sensitivity and can lead to false negative results. In automatic detection algorithms, knowledge about breast density can also be useful for setting an appropriate threshold. It is impossible to produce satisfactory classification results by knowledge of overall intensity because exposure and breast volume are different. Because of that we observe breast density as a texture classification problem. In this paper we propose feature selection process based on Haralick and Soh feature set with optimization for k-nearest neighbor classifier. Feature selection was done by individual feature ranking, using linear forward selection and finally using wrappers. The best feature selection results were obtained using wrappers. The improvement on overall classification is 6.8% in comparison to the classification without feature selection on the same dataset.