Particle Filtering Methods for Subcellular Motion Analysis

Advances in fluorescent probing and microscopic imaging technology have revolutionized biology in the past decade and have opened the door for studying subcellular dynamical processes. However, accurate and reproducible methods for processing and analyzing the images acquired for such studies are still lacking. Since manual image analysis is time consuming, potentially inaccurate, and poorly reproducible, many biologically highly relevant questions are either left unaddressed, or are answered with great uncertainty. The subject of this thesis is particle filtering methods and their application for multiple object tracking in different biological imaging applications. Particle filtering is a technique for implementing recursive Bayesian filtering by Monte Carlo sampling. A fundamental concept behind the Bayesian approach for performing inference is the possibility to encode the information about the imaging system, possible noise sources, and the system dynamics in terms of probability density functions. In this thesis, a set of novel PF based methods for subcellular motion analysis is developed. The applicability of these methods for robust and accurate detection and tracking of large numbers of small objects in 2D and 3D image sequences obtained by fluorescence microscopy imaging as well as for dynamics analysis using kymoghraphs is demonstrated and evaluated. The proposed algorithms have been tested on synthetic image data as well as on real time-lapse fluorescence microscopy data acquired for studying the dynamics of three different types of intracellular objects: microtubules, vesicles, and androgen receptors. Compared to existing approaches the reported methods are substantial improvement for detection and tracking of large and time-varying numbers of subcellular objects in image data with extremely low signal-to-noise ratio.