A CAD System for Identification and Classification of Breast Cancer Tumors in DCE-MR Images Based on Hierarchical Convolutional Neural Networks

In this paper، we propose a computer aided diagnosis (CAD) system based on hierarchical convolutional neural networks (HCNNs) to discriminate between malignant and benign tumors in breast DCE-MRIs. A HCNN is a hierarchical neural network that operates on two-dimensional images. A HCNN integrates feature extraction and classification processes into one single and fully adaptive structure. It can extract two-dimensional key features automatically، and it is relatively tolerant to geometric and local distortions in input images. We evaluate CNN implementation learning and testing processes based on gradient descent (GD) and resilient back-propagation (RPROP) approaches. We show that، proposed HCNN with RPROP learning approach provide an effective and robust neural structure to design a CAD base system for breast MRI، and has potential as a mechanism for the evaluation of different types of abnormalities in medical images.