Automated Breast Cancer Image Classification Based on Integration of Noisy-And Model and Fully Connected Network

In this paper, we proposed an automated pathological image classification approach for supporting breast cancer (BC) diagnosis, e.g., BC image classification for categories of normal, benign, in-situ and invasive. The proposed model is consist of two components: first, a dual path network (DPN), which is a deep convolutional neural network used to convert R.G.B. features of the given input image into a probability map of each possible category; and second, a integration of a noisy-and model and a fully connected neural network is used as a classifier, which takes both global and local features into account in order to achieve a better performance. Based on 10-fold cross validation using the given training set, the accuracy of the proposed approach was \({\sim }91.75\%\). The accuracy on the test set provided by the contest, the accuracy was \({\sim }64.00\%\).