Chapter 6 - Mitosis detection in biomedical images

Abstract

Mitosis is an important stage in the life cycle of cells. The analysis of mitosis can provide important information for the study of cell behaviors during cell culture. In addition to being used to analyze the proliferation of cells, mitosis detection can be used to improve the accuracy of cell-tracking results. In this chapter, we will give a detailed description of the mitosis detection problem, and then introduce some advanced research in mitosis detection areas in recent years. In Section 1 of this chapter, we introduce the mitosis process during the cell life cycle and then give a definition of the mitosis detection problem from the perspective of computer vision. In Section 2, we introduce some common microscopy imaging methods for mitosis detection. Various microscopy imaging modalities can lead to different techniques used for mitosis detection. It is essential to have some understanding of these microscopy imaging methods. Based on these techniques, mitosis can be detected from a single static image or a sequence of dynamic images. Computer vision and machine-learning techniques used to deal with both situations are further described in Section 3, and then we divide the mitosis detection methods into four categories: traditional tracking-based methods, tracking-free methods on static images, hybrid methods with short-term tracking and modeling of the probability graph model, and finally, the recently proposed deep learning-based methods. For mitosis detection on static microscopy images, we focus the description of some popular image feature extraction methods, including the application of some classic features such as GIST and SIFT. We further introduce the recent research findings for the dynamic modeling of the mitosis process. In some applications, this dynamic information can provide important information that helps to improve the accuracy of mitosis event detection. The dynamic information can be extracted with cell-tracking methods or spatial-temporal features, modeled with a probability graph model or an RNN model. We also introduce the application of deep learning methods to combine the spatial appearance feature and the temporal dynamic for mitosis detection tasks. Finally, in the last section, we summarize these existing methods and discuss the potential directions for the mitosis detection problem in future research.