Mitosis detection plays important role in many bio-medical applications such as medical diagnosis and drug development, by providing important information about cell division behaviors. Phase contrast microscopy imaging provides an advantageous tool for real time observation of cells without cell damage caused by staining. Spatial visual information, together with temporal dynamic information captured by continuous microscope imaging, could be used for automatic detection of mitotic events among whole-slide images with computer vision and machine learning techniques.
The aim of this challenge is to provide a common benchmark for the evaluation of the mitosis detection algorithms in a new dataset of whole-slide phase contrast time-lapse microscopy images.
Mitosis detection aims to determine the presence of mitosis in the microscopy image, and then locate the spatial and temporal location of mitotic cells across the whole-slide temporal image sequence. Unlike mitotic detection in static images, mitosis detection in phase contrast image dataset need to take both spatial and temporal information into consideration for more precise detection results.
Mitosis detection in phase contrast time-lapse microscopy image has made great progress in recent years of research. Compared with traditional hand-crafted feature based methods, deep learning based method received extensive attention. Several deep learning based methods have been proposed for mitosis detection task in recent years, and they are commonly evaluated on some popular datasets such as the C3H10 and C2C12 dataset. These datasets contain relatively few annotated mitotic events, which could not adequately address the need for big data in deep learning methods. However, since the acquirement of mitosis annotation takes much expertise, it is expensive to get large scale of annotated dataset. In order to promote the development of mitosis detection algorithms, we propose a competition with a dataset that contains more annotated mitotic events than previous datasets.