Classification of Histopathology Images of Breast into Benign and Malignant using a Single-layer Convolutional Neural Network

Breast cancer is known as the second most prevalent cancer among women worldwide, and an accurate and fast diagnosis of it requires a pathologist to go through a time-consuming process examining different captured images under varying magnifications. Computer vision and machine learning techniques are used by many scholars to automate this process and provide a faster and more accurate diagnosis of such cancers, but most of them have utilized handengineered feature descriptors to classify the type of images (whether benign or malignant) based upon. Deep learning techniques have made a significant progress in the world of pattern recognition, image classification, object detection, etc. Convolutional Neural Networks (CNN) -- a special kind of deep learning methods -- are best known for identifying patterns in the images; they try to represent an abstract form of images containing the most salient information needed for distinguishing them from different similar-looking images. The main aim of this paper is to employ CNN for the task of breast cancer classification given an unknown image of the patient for an accurate diagnosis. A new network design is proposed to extract the most informative features from a collection of histopathology images provided by BreakHis database of microscopic breast tumor images. The experimental results carried on 1,995 histopathological images (with a 40× magnifying factor), demonstrated an improved accuracy compared to some prior works, and a comparable performance regarding one of the previous works.