Exploiting rotation invariance with SVM classifier for microcalcification detection

In previous work we developed a support vector machine (SVM) approach for detection of microcalcifications (MCs) in mammogram images, which was demonstrated to outperform several existing methods for MC detection in the literature. In this work, we explore whether we can further improve the performance of the SVM detector by exploiting the fact that MCs are inherently invariant to their spatial orientation in a mammogram image. We consider two different techniques for incorporating invariance into SVM, of which one is virtual support vector SVM (VSVM) and the other is tangent vector SVM (TV-SVM). In the experiments these techniques were tested on a database of 200 mammograms containing a total of 5,211 MCs. The results show that both techniques can improve the performance in discriminating MCs from the image background, and TV-SVM achieved the best performance.