Fusion of Deep Features for Classification of Breast Cancer Using Multi-Deep CNNs

Abstract

AbstractBreast cancer remains a deadly disease that frightens women in several parts of the world. At the same time, deep learning becomes the widely used and fast-growing area of traditional machine learning. The work experiments a newer computer-aided diagnosis (CAD) tool that comprises of feature extrication and classification through deep learning for assisting radiologists in breast cancer classification in mammogram images. And this is done by three different experimentations for determining the optimum way of robust classification. Herein, the first one makes use of pretrained Deep CNNs namely AlexNet, GoogleNet, Res-Net50, and DenseNet121. The second one is based on the experimentation of extracting features using the Deep CNNs and applied to a support vector machine (SVM) model. The last one is based on fusing the deep features for designing a robust classification framework. All these experimentations are evaluated using MIAS database. And finally, the results reveal that the fusing of deep features enhanced the classification performance of SVM, i.e., this deep feature fusion (feature set_3) with SVM provides a maximum classification accuracy of 96.739% than other approach.KeywordsConvolution neural networksSupport vector machineDeep learningTransfer learningGaussianBreast cancerMammograms