Survey on Mitosis Detection for Aggressive Breast Cancer from Histological Images

Abstract

The mitotic count is a relevant factor for grading invasive breast cancer. Since it is subject to human prone error, requires more time for completion and the nuclei look similar during all stages of mitosis, automatic detection of mitosis is a good solution to overcome these problems. In this paper, the top methodologies used for mitosis detection are analyzed. Some of them were a part of challenging competitions conducted worldwide. Analysis of the result shows that top approaches, either implemented Random Forest (RF) classifier exploiting intensity feature or used deep learning methods like Convolutional Neural Network (CNN) to give out the best results. It was also found that the ensemble classifiers gives better performance. A preliminary experiment conducted on cascaded RF and Artificial Neural Network (ANN) results in better accuracy than individual classifiers.