Automated Diagnosis of Breast Cancer from Mammogram Using Wavelet, Curvelet Features, and Convolutional Neural Network

Abstract

Breast cancer is the utmost generally occurring cancer in women and the second most communal cancer. The ground truth standard used in real-time clinical application for the diagnosis is a mammogram. A novel approach is projected in this paper for the automated diagnosis of breast cancer from mammogram images composed from the MIAS data set using curvelet/wavelet transform-based features and a convolutional neural network. The following sequences of operations are involved, namely pre-processing, application of curvelet/wavelet transform, statistical and gray level co-occurrence matrix-based features extracted from curvelet/wavelet coefficients followed by a selection of highly discriminative features by statistical p-test. Initially, pre-trained models VGG16 and VGG19 are used for classification, and Deep convolutional neural network architecture is constructed for which feature matrix is given as input. Pretrained models are used for classification using the concept of transfer learning. The constructed architecture hyperparameters are adjusted, and the highest classification precision of 93% is achieved. The obtained results outperform the state of art methods available in the state of art.