The Tracking & Analysis of Cell Movement with Cell-Derived Active Contours & Generative Graphical Models

Abstract

There has been a continuous effort (Li et al., 2009; Al-Kofahi and Lassoued, 2010; Al-Kofahi et al., 2011; Ambühl et al., 2012; Meijering, 2012; Maška et al., 2014) to provide tools for large-scale automated analysis of live cell morphology in vitro, as this very important for studying biological development and disease. In this thesis, we outline a set of theoretical and practical tools that we have developed for this purpose. In chapters 1-5 we present a new cell-derived active contour (CDAC) method for cell tracking that achieves higher tracking accuracy using knowledge about cell movements. This knowledge consists of the way cells move by adhering to the substrate and propelling their cell membranes in the direction normal to the edge. This allows use of fast optimal search methods. Our method is designed for phase contrast microscopy, which allows access to a wider variety of experimental data. It can operate under conditions of low illumination. This is important for live-cell imaging as high illumination can damage cells. Additionally, our method demonstrates a higher robustness to noise than many previous methods. We then develop a statistical framework for cell shape analysis. Due to the large dimensionality of the space of cell shapes, we first use a neural network model to extract a small set of relevant features. This allows efficient learning and this makes our method distinct from previous methods (such as kernel density estimation). We then develop a rotation-invariant framework - a property that has not been incorporated in previous cell shape model studies and is a limitation of other models - to construct a hierarchical Bayesian generative model to learn the distribution of cell shapes. We show that this new property allows, for the first time, deriving features such as elongation or cell asymmetry directly from the data without any prior assumptions. In this work we maintain a close relationship between experiment and theory. The problem of calculation of cell movement indices from experiment is studied and related to theoretical models. We also use theoretical models to validate our morphology analysis framework, showing good statistical agreement with natural cells.