Combined Use of Local and Global Features for Classification of Breast Lesion Using DCE-MRI Images

Abstract

Recently, the use of dynamic contrast enhanced-magnetic resonance imaging (DCE-MRI) technique is widely used to detect and diagnose breast cancer. This technique has shown to be very useful particularly in screening women with high risk for breast cancer, as well as assessing the potential effects of new therapy. Thus, the aim of the present study is to appraise the efficacy of combined employment of global and local features in discriminating malignant and benign lesions. A dataset of one hundred and twenty one (121) DCE-MRI investigations was assembled and used. Out of that number, fifty (50) were biopsy-proved malignant tumors and seventy-one (71) were benign. Firstly, the suspicious mass regions were automatically detected and segmented with 3D region growing algorithm. Meanwhile, Local and global features were used. Thereafter, sequential floating forward selection method (SFFS) and support vector machine classifier (SVM) were used for classification. The overall classification performance of different kind of features were evaluated via receiver operating characteristic (ROC) analysis in a 3-fold cross validation scheme. It was observed that global feature produced classification accuracy of 84.32 % followed by local feature with accuracy of 85.95 %. When the local and global features were combined, the classification accuracy increased to 94.36 %. Based on the obtained results, this study has demonstrated that the combined use of local and global features could effectively function as a better indicator in differentiating malignant and benign tumors.