Keywords
Breast Cancer, Artificial Intelligence, Deep Learning, Mitosis Detection.
Methods
Several deep learning models were used to detect and classify mitotic cells. First, all the nuclei of cells in the histopathological images of the breast tissue were segmented using U-Net architecture. The outputs of the nuclei detection model were regarded as ROI for mitosis detection. After detecting nuclei, VGG-11 architecture was used to make predictions for each nucleus region to determine whether the nucleus is mitotic or non-mitotic. A unique dataset was created using 4794 mitotic and 4794 non-mitotic cell nuclei.
References
[1] Sunil R. Lakhani, Ian O. Ellis, Stuart J. Schnitt, Puay Hoon Tan, Marc J. van de Vijver (Ed.). (2012). WHO Classification of Tumours of the Breast (4th ed.) International Agency for Research on Cancer (IARC), Lyon, France: pp 19-20