Multiclass Classification of Breast Cancer in Whole-Slide Images

Abstract

Breast cancer is one of the leading cause of cancer-related death worldwide. During the diagnosis of breast cancer, the histopathological assessment of Haemotoxylin and Eosin(H&E) stained slides provides important clinical values. By applying computer-aid diagnosis on whole-slide image(WSI), the efficiency and consistency of such assessment could be improved. In this paper, we propose a deep learning-based framework that classifies H&E stained WSIs into regions of normal tissue, benign lesion, in-situ carcinoma and invasive carcinoma. The framework utilizes both microscopy images and WSIs to train a patch classifier in two stages. The underlying classifier is based on Inception-Resnet-v2. This framework won both parts of the ICIAR2018 Grand Challenge on Breast Cancer Histology Images [4] competition, achieved a part A multiclass accuracy of 87% and part B score of 0.6929.