Breast density classification using Laws' mask texture features

Abstract

Characterisation of tissue density is clinically significant as high density is associated with the risk of developing breast cancer and also masks lesions. Accordingly, in the present work, PCA-kNN and PCA-NN based computer aided diagnostic (CAD) systems for breast tissue density classification have been proposed. The work has been carried out on the MIAS dataset. Five statistical texture features mean, standard deviation, entropy, kurtosis and skewness are evaluated from Laws' texture energy images resulting from Laws' masks of lengths 3, 5, 7 and 9. Principal component analysis is then applied to these texture feature vectors for feature space dimensionality reduction. The kNN classifier and the NN classifier are used for the classification task. The highest classification accuracy of 95.6% is achieved by using the first 8 principal components computed from texture features derived from Law's mask of length 5 for k = 8 using the kNN classifier.