Convolutional Neural Network – Regularized Extreme Learning Machine with Hyperbolic Secant for Breast Cancer Segmentation and Classification

Early detection of breast cancer by mammogram images is essential to reduces the mortality rate of women and allow for appropriate treatment. Deep Learning (DL) algorithms are generally utilized for feature extraction and have shown effective performance. Though, these features are performed well in certain cases because of redundant and irrelevant data. To overcome this, in this manuscript developed the Convolutional Neural Network - Regularized Extreme Learning Machine (CNN-RELM) with Hyperbolic Secant (HS) activation function for precise breast cancer segmentation and classification. Initially, the images are segmented by using the Optimized Region Growing (ORG) algorithm which effectively segments the images to differentiate the different classes of tumor regions. The CNN layers extracted both low-level and high-level features and reduced dimensions while retaining significant features. The RELM includes the regularization term for enhancing the generalization ability and handles the noise data. The HS activation function process based on hyperbolic secant function and offers smooth activation, good gradient flow and the network generalization ability. The developed RELM-CNN with HS method obtained 99.87% accuracy on MIAS dataset while comparing with conventional algorithms.