Subcellular Structure Modeling And Tracking For Cell Dynamics Study
Abstract
The introduction of sensitive and fast electronic imaging devices and the development of biological methods to tag proteins of interest by green fluorescent proteins (GFP) have made a full understanding of live cell dynamics achievable. With the latest hardware technology, such as high speed laser scanning confocal Microscopy (LSCM), it has now become critical to develop automatic quantitative data analysis tools to keep pace with and to fully exploit the functionalities of state-of-the-art hardware. One task of such tools is the motility analysis of subcellular structures. This dissertation provides a series of computational approaches for studying subcellular structure motility.
Firstly, a semi-automatic single object tracking approach using sequential Monte Carlo (SMC) method is developed. To achieve reliable tracking, a flow of criterion for object feature selection, matching,
and evaluation criteria are designed: a grid-based minimum variance (GMV) feature selection rule, a mean minimum to maximum ratio (MMMR) similarity measurement, and the feature evaluation tests by feature convergence ratio (FCVR) and feature consistence ratio (FCSR).
Secondly, to handle complex scenario of multiple interacting subcellular structure motion, we apply reversible jump Markov chain Monte Carlo (RJMCMC) method to sample the distribution of the dimension varying joint state which is the combination of the states
of multiple individual subcellular structures. Five RJMCMC moves are
constructed, including object appear move, disappear move, update
move, height swap move, and identity swap move. The evolution of each
individual state in the joint state is sampled by the update move. In
order to deal with the random appearance locations of subcellular
structures, a marker residual image guided appearance model is
proposed to detect the newly appearing object, and the appear move
and the disappear move are applied to generate samples resulting
from the new object appearance. To prevent the RJMCMC sampling from
being trapped at the local maxima, the identity swap move is also
constructed. The proposed RJMCMC SMC tracking approach is applied to
numerous time-lapse LSCM video sequence tracking in both 2D+T and
3D+T domains.
Finally, from a perspective different from the SMC framework, we model the multiple object tracking as a bipartite graph matching problem between the consecutive image frames. To save the possible high cost of graph matching, a Markov chain Monte Carlo data association (MCMCDA) method with deletion move, switch move, and
addition move is developed to approximate the optimal solution.