Mass Detection in Mammograms
Ping‐Sung Liao, Shu‐Mei Guo, Nan-Sue Yu, Li-Chun Chen, San-Kan Lee, Chein‐I Chang · 2012
Many research efforts devoted to development of computer-aided-diagnosis (CAD) systems for mass detection have been focused on feature extraction/selection which is a crucial step in success of a CAD system. This paper investigates and evaluates many feature extraction techniques developed for mass detection in mammgrams. In particular, two major techniques, texture spectrum and texture feature coding method are explored and a new feature extraction descriptor, called just noticeable difference (JND) is introduced. In order to improve accuracy for mass detection, the principal components analysis (PCA) and a new proposed genetic algorithm (GA) are used to select an optimal set of features that are fed to two neural network classifiers, backpropagation neural network (BPNN) and probabilistic neural network (PNN) for classification. The experimental results show that the proposed genetic algorithm outperforms the PCA in feature selection. The results also show that the best classification can be obtained by combining the proposed GA with a PNN classifier.