Breast cancer classification using covariance description in Riemannian geometry
Cyrus Avaznia, Seyyed Hamed Naghavi, Mohammad Bagher Menhaj, Hamed Talebi · 2017
In this paper, we analyze the performance of Minimum Distance to Riemannian Mean (MDRM) and Tangent Space Linear Discriminant Analysis (TSLDA) to classify the medical images suspicious to be malignant. MDRM and TSLDA have been previously used as a new classification framework for brain-computer interface (BCI), but their application to classify mammogram images is a novel idea which is considered here. We first segment breast masses in the images by wavelet analysis and genetic algorithm. Then, a covariance descriptor is used to extract the features of the breast masses in these images. And Finally, the classification is employed by applying MDRM & TSLDA on the extracted features to anticipate the class of raw suspicious images. To illustrate the high accuracy (specificity) of the proposed classifiers, the results of NLSVM has been provided besides the results of MDRM and TSLDA.