Breast Cancer Histology Image Classification Based on Deep Neural Networks

Abstract

Automatic histopathology image recognition system plays a key role in speeding up diagnosis and reducing the error rate. In this work, we propose a new histopathology image recognition scheme. First, hybrid Convolutional Neural Network (CNN) architecture is designed based on GoogLeNet to merge more key information in decision. Second, bagging technique and hierarchy voting tactic are executed to reduce generalization error and improve performance. At last, transfer learning and data augment strategies are employed to address the limitations of the small amount of data. The classification accuracy of our adopted method is 87.5% for four class, which far outperforms the existing state-of-the-art.