Mitosis Detection in Breast Cancer by Inference of Segmentation and Bag of Features

Breast cancer is one of most prevalent form of cancers in females worldwide [1]. Normal as well as cancerous cells multiply through a biological process called as mitosis. Mitosis count is a paramount feature in determining the aggressiveness or growth rate of cancer. The cells that undergo mitosis exhibit certain stages due to which they can be distinguished from non-mitotic cells. Usually pathologists identify mitotic cells in histological images, which is a laborious and timeconsuming task. In this paper we have developed a procedure to automate this process. The proposed methodology constitutes of segmentation, feature extraction and classification using support vector machine. Classification of cells is performed based on features extraction of statistically segmented sub images. Bag of features has been adapted as feature extraction technique, specifically utilizing constraints of intensity, shape, texture and curvity of the cells. We have utilized MITOS-ATYPIA 14 data-set [2] for our research. System is verified against the ground truth using two evaluation parameters sensitivity and F1 score. Adopting the proposed methodology produced the adequate results in specifying the sensitivity of mitotic cells, with sensitivity of 100% and F1 score 71.4%.