Fully Convolutional Neural Networks for Tissue Histopathology Image Classification and Segmentation

Abstract

Analyzing tissue histopathology image is a new research area of the computer vision. It is getting more and more attention recently. Many pathologists and researchers have gained remarkable achievements in this area. However, most of previous works focused on improving the accuracy rather than the efficiency. In this paper, we propose a novel classification and segmentation framework based on fully convolutional neural networks, which can largely speed up large histopathology image analysis without damaging the accuracy too much. The method is validated through experiments on the Hematoxylin and Eosin (H&E) images of stomach tissues collected from real world. The experimental results show that the proposed method achieves comparable accuracy in both classification and segmentation tasks, along with a 16 times faster speed than the tested state-of-the-art methods.