Mitotic cells detection in H&E-stained breast carcinoma images

Breast cancer is the most common cancer occurring in women, and is the second leading cause of cancer related deaths in women. Grading of breast cancer is carried out based on characteristics such as the gland formation, nuclear features, and mitotic activities, all of which need to be correctly detected first. In this paper, we proposed a system to detect mitotic cells from H&E-stained whole-slide images of breast carcinoma. The system consists of three stages, namely superpixel segmentation to group similar pixels into superpixel regions, blob analysis to separate the cells from the tissues and the background, and shape analysis and classification to distinguish mitotic cells from non-mitotic cells. The proposed system, with the histogram of oriented gradients (HOGs) and Fourier descriptor (FD) as features, is able to detect mitotic cells reliably, with more than 90% true positive rate, true negative rate and overall accuracy.

Automated summary

In this paper, a system for detecting mitotic cells in breast carcinoma using image processing and morphological analysis is proposed, achieving accurate detection and classification of cells.