Nuclear Atypia Grading in Histopathological Images of Breast Cancer Using Convolutional Neural Networks

Early detection of breast cancer can efficiently increase the success of treatment. One of the criteria for diagnosis and grading of breast cancer is nuclear atypia. Manual Grading of histopathological images is a subjective and time consuming task. Therefore, it's necessary to provide an automatic diagnostic system for grading histopathological images. In this paper, we present an automatic diagnostic system that classify histopathological images based on nuclear atypia criterion. According to a recent success of Conovolutional Neural Networks (CNNs) in image classification, in this paper, a CNN-based method has been used. An image augmentation method are applied to the images. Then they are processed before entering the network to better differentiate the colors. The images are processed in L*a*b* color space and finally images are entered to the proposed network for nuclear atypia grading. Simulation results and comparison to other related works show the efficiency of the proposed system.