Classification of Breast Cancer Subtypes Using Deep Learning Methods

Abstract

This paper describes a study on the classification of breast cancer subtypes using deep learning methods. BRACS dataset containing RoI images colored with hematoksilen and eozin was used. In the data preprocessing stage, pixels that do not contain cancer cells were temporarily colored black to make the cancer cells more clearly visible. Then, new datasets were created by extracting 128, 256 and 512 pixel images from the regions where cancer cells are dense. In the classification phase, deep learning models were used and their success was measured and compared according to accuracy metrics.