Accuracy Improvement in Binary and Multi-Class Classification of Breast Histopathology Images

Abstract

Breast cancer is widely common in women. In order to recognize this type of cancer, experienced pathologists must evaluate cell shapes in breast histopathology images in different magnification levels. However, there are few pathologists per population in many countries, and there is always a possibility of human error. This study represents a deep transfer learning-based model that can improve some of the current state-of-the-art results in both binary and multi-class classification. The presented model has been examined in magnification-dependent and magnification-independent modes. As for multi-class classification, the proposed system grants 94-97% accuracy. In binary classification, the proposed system provides up to 99% accuracy. The results show higher accuracies than past studies in the classification of breast histological images in some cases with the same algorithm and setting for all the measurements.