Breast Cancer Histology Image Classification using Deep Learning

The breast cancer histology image classification task is to classify images into four categories of normal, benign, in situ carcinoma, and invasive carcinoma. A common challenge of the breast cancer histology image classification, also in other medical domain, is a lack of sufficient data. To address the limitations of the small amount of data, we applied additional patch extraction (APE) of whole-slide image (WSI) as dataset extension approach. Besides, to improve the classification accuracy, we proposed to use the test time augmentation (TTA) technique (with horizontal/vertical flipping, and ±90 degree rotation) upon the original convolution neural network (CNN) so that model is able to make a better decision on several breast cancer testing images instead of single prediction. Experiments was done by applying APE on ICIAR 2018 dataset to get the extended one (ICIAR-EXT) which is publicly available at https://github.com/canhnp/ICIAR-EXT. Moreover, when applying TTA technique with CNN on ICIAR-EXT at testing state, the accuracy has achieved a satisfactory result of 78% for 4-class breast cancer classification for 100 unlabeled ICIAR 2018 test set.